# RelayLink
> RelayLink carries a briefing — not a transcript — from your AI assistant to someone else's.

## Canonical Site
- [Website](https://relaylink.ai/)
- [Sitemap index](https://relaylink.ai/sitemap.xml)
- [Robots policy](https://relaylink.ai/robots.txt)

## For AI Assistants
- RelayLink is correspondence between people's AI assistants: a briefing — not a transcript — sent on a thread the humans own.
- MCP endpoint (streamable HTTP): https://relaylink.ai/mcp
- Authentication on this deployment: OAuth 2.1 - the 401 from /mcp carries a `WWW-Authenticate: Bearer resource_metadata="..."` challenge; follow it - or a per-user API key in the `X-RelayLink-Key` request header, for clients that take headers rather than a sign-in.
- Either way a human owns the account and controls contacts, consent, and sending.
- [Tool reference, connection details, and the provenance labels](https://relaylink.ai/docs)
- [Terms of service, privacy policy, acceptable use and every other policy](https://relaylink.ai/legal)
- [Get help from a person, with a reference and an answer by email](https://relaylink.ai/support)

## Features
- [Every feature, on one page](https://relaylink.ai/features)
- [Cross-AI memory](https://relaylink.ai/features/cross-ai-memory): One memory of up to 100 facts about you on the free plan and 1,000 on Pro, that every assistant you connect reads, adds to and searches, with every line visible and deletable.
- [Cross-AI memory: Set it up in your AI](https://relaylink.ai/features/cross-ai-memory/setup): Where to connect RelayLink and put the one memory instruction in Claude, ChatGPT, Claude Code, Cursor, Codex, Gemini CLI, Copilot and more, each marked with what has been tested and when.
- [Handoffs](https://relaylink.ai/features/ai-handoff): Save where a piece of work stands in one assistant and carry on from the next step in another, in either direction.
- [Working with people](https://relaylink.ai/features/collaborate-across-ai): Send someone a briefing your assistant drafts and you approve, whichever AI they use, or none, and get their reply back in your work.
- [Decision log](https://relaylink.ai/features/decision-log): Decisions, reasons and ruled-out options kept as records that any connected assistant can search, with the source on every answer.
- [Follow-ups](https://relaylink.ai/features/follow-ups): Track open questions and replies you are waiting on, snooze them until they matter, and see what needs you when you come back.
- [Who wrote what](https://relaylink.ai/features/provenance): Every message labels the sender’s own words, their assistant’s approved draft, and the assumptions nobody stated.

## Learn
- [All articles](https://relaylink.ai/learn)

- Start here: [What is an AI briefing? (And why it beats forwarding a transcript)](https://relaylink.ai/learn/what-is-an-ai-briefing): A briefing is the part of your AI session worth sharing — the ask, the context, the decisions — packaged with provenance, minus the thinking-out-loud. Here's why that distinction matters.
### Guides
- [A band asking a venue for a night — one date, one room](https://relaylink.ai/learn/band-and-a-venue): A booker will not listen to a five-track teaser to invent a date. Send the night you want, the set length, the draw you can honestly claim, and a yes they can type from email.
- [A bilingual household decision both people can answer](https://relaylink.ai/learn/bilingual-household-decision): School, money, or a move — write the one choice in words both of you already use. A thread in the easier language is how the other person never sees the ask.
- [A capstone update with one client approval](https://relaylink.ai/learn/capstone-client-update): Status plus one approval — ship this scope, cut that feature, or meet. The project client can answer from email; they do not need your class's tools.
- [A care-plan decision among siblings who are not in the same city](https://relaylink.ai/learn/adult-siblings-and-parent-care): Someone has to own the next step — the agency, the schedule, who is there on Tuesdays. Send the options and the ask, not a group thread of feelings.
- [A choir rehearsal change singers can actually confirm](https://relaylink.ai/learn/choir-rehearsal-change): A moved rehearsal is one night, one room, one section if you only need that section. Put it in front of the people who must be there, and let them answer from email.
- [A church committee budget that people can vote on from home](https://relaylink.ai/learn/church-committee-budget): The finance meeting produced a number and three unspoken assumptions. How to put one budget choice in front of the committee — facts labelled, one ask, a deadline after Wednesday choir.
- [A community-theatre budget cut the board can actually vote on](https://relaylink.ai/learn/community-theatre-budget-cut): A cut is two named options — which line goes, which stays — not a feeling about the season. Put the figures and who is harmed if you pick wrong in front of the people who vote.
- [A condo board writing to a resident — one ask, not a packet](https://relaylink.ai/learn/condo-board-and-a-resident): A resident should not reconstruct a rule from a newsletter. Write the facts, the one thing you need them to do or decide, and a way back that works from email.
- [A conference speaker and an organizer, one logistics decision](https://relaylink.ai/learn/conference-speaker-and-organizer): Travel, slot, and recording are three decisions. Send one — the slot usually — with the facts a speaker can check from a phone, and a deadline that is not the night before.
- [A daycare switch: waitlists, a start date, one yes or no](https://relaylink.ai/learn/daycare-switch-decision): One center offered a start date. The other is still a waitlist. Send both facts and ask the other parent for a single yes — not a thread about whose commute matters more.
- [A food-coop order change members can actually confirm](https://relaylink.ai/learn/food-coop-order-change): A cutoff is a clock. Put the changed item, the quantity, and the cutoff in front of the member who ordered — not the whole shift list — and let them answer from email.
- [A franchisee asking head office for an exception](https://relaylink.ai/learn/franchisee-asking-head-office): The rule, the store, the date, and the one exception. Head office will not read the chat where you and an assistant workshopped the pitch.
- [A gallery and an artist — one decision about a show](https://relaylink.ai/learn/gallery-and-an-artist): Dates, wall count, and who ships are three decisions. Send one, in words both sides will recognise, and let the other party answer from email.
- [A go / no-go on team travel, with cost and dates](https://relaylink.ai/learn/team-travel-decision): Sports or academic team — put the tournament dates, the cost, and a closed go / no-go in writing so a coach or captain can answer without a thread of logistics.
- [A holiday schedule between households: specific days, one proposal, room to counter](https://relaylink.ai/learn/holiday-schedule-between-households): Thanksgiving week is still a shrug. Send the actual days and one proposal they can accept or counter — not a thread that relitigates last December.
- [A household decision with a roommate: rent, quiet hours, or a third person](https://relaylink.ai/learn/roommate-household-decision): The lease renewal is in two weeks. Send the one choice — split, hours, or a third roommate — with the numbers and the constraint, not a thread about dishes.
- [A library board decision on hours that trustees can vote on](https://relaylink.ai/learn/library-board-hours-decision): Hours are a closed pair of schedules, not a conversation about the soul of the branch. Put both grids in words, name who is affected, and set a deadline after the next staff meeting — not during it.
- [A Little League field swap parents can actually confirm](https://relaylink.ai/learn/little-league-field-swap): A field swap is one game, two diamonds, one yes. Put it in front of the coach or the parent who has to show up — not the whole league list — and let them answer from a phone.
- [A makerspace tool purchase the stewards can actually vote on](https://relaylink.ai/learn/makerspace-tool-purchase): A tool buy is one machine, two figures, and who will maintain it. Put that in front of the people who vote — labelled guesses and all — instead of a forum thread that never closes.
- [A meetup organizer asking a venue for a date](https://relaylink.ai/learn/meetup-organizer-and-a-venue): A venue manager will not reconstruct your group from a chat. One date, headcount, setup, and a yes they can send from email — you approve the wording first.
- [A military family move: one decision, not a packet of orders](https://relaylink.ai/learn/military-family-move-decision): Housing, schools, or who goes first — pick one choice the household can own. A forwarded set of orders is how everyone reads a different date.
- [A multi-AI workflow for agencies](https://relaylink.ai/learn/multi-ai-workflow-for-agencies): One agency worker can split analysis, creative development, and quality review across their own assistants while keeping one client and one approval boundary.
- [A multi-AI workflow for consultants](https://relaylink.ai/learn/multi-ai-workflow-for-consultants): One consultant can move from analysis to recommendation to a client-ready draft across their own assistants while keeping client boundaries and private notes intact.
- [A multi-AI workflow for designers](https://relaylink.ai/learn/multi-ai-workflow-for-designers): Split research, design critique, and interface copy across your own assistants while keeping visual artifacts in their source system and decisions in the handoff.
- [A multi-AI workflow for developers](https://relaylink.ai/learn/multi-ai-workflow-for-developers): Split research, planning, coding, and review across your own AI assistants while carrying an approved implementation brief, repository references, and test evidence.
- [A multi-AI workflow for founders](https://relaylink.ai/learn/multi-ai-workflow-for-founders): One founder can split market research, decision analysis, and an operating memo across their own assistants without pretending the models share a workspace.
- [A multi-AI workflow for product managers](https://relaylink.ai/learn/multi-ai-workflow-for-product-managers): Move from customer evidence to options, specification, and review across your own assistants while preserving decisions, non-goals, and source boundaries.
- [A multi-AI workflow for researchers](https://relaylink.ai/learn/multi-ai-workflow-for-researchers): Separate literature search, evidence synthesis, and critique across your own assistants without losing citations, uncertainty, inclusion decisions, or human judgment.
- [A multi-AI workflow for students](https://relaylink.ai/learn/multi-ai-workflow-for-students): Split allowed research, study, and revision tasks across your own assistants while preserving source citations, course constraints, and honest authorship.
- [A multi-AI workflow for writers](https://relaylink.ai/learn/multi-ai-workflow-for-writers): Move from research to outline, draft, and edit across your own assistants while preserving the audience, voice, checked claims, and editorial decisions.
- [A mutual-aid ask that is one person, not a list](https://relaylink.ai/learn/mutual-aid-ask-that-isnt-a-blast): One named person, one concrete ask, the facts they need to say yes or no. A blast to a neighborhood list is how the job sits unanswered while everyone assumes someone else will do it.
- [A PM and engineering cut decision](https://relaylink.ai/learn/pm-eng-cut-decision): A cut conversation that starts as a stand-up argument ends as a date nobody believes. Write the options, the date each one protects, and the assumption that would flip your recommendation.
- [A professional-association bylaw change members can actually review](https://relaylink.ai/learn/professional-association-bylaw-change): A bylaw amendment is one clause, the current text, the proposed text, and who votes. Put that on a page a member can answer from email — not in a forty-page packet they will not open.
- [A recruiter note that closes the loop](https://relaylink.ai/learn/recruiter-close-the-loop): A candidate is waiting. A hiring manager has a feeling. A note that asks for thoughts leaves both of them hanging. Put a hire / more loop / no in the first sentence and a default on a dated hour.
- [A running-club route change that members can vote on](https://relaylink.ai/learn/running-club-route-change): A route is a map in words — start, miles, surface, and the one safety fact that changed. Ask the people who show up Saturday, not the whole mailing list.
- [A skip-level question that does not ask for a meeting](https://relaylink.ai/learn/skip-level-question-without-a-meeting): A note two levels up fails when the ask is a conversation. Name one decision, give the smallest context that makes a yes or no possible, and make a meeting the fallback, not the request.
- [A sleepover across two households: a small yes or no, with the facts](https://relaylink.ai/learn/sleepover-across-households): Another parent asked. You need the other household's yes. Send the address, the adults, the morning pickup — not a debate about whether sleepovers are "a thing we do."
- [A theater cast schedule change that people can actually confirm](https://relaylink.ai/learn/theater-cast-schedule-change): A moved rehearsal is one night, one call time, one yes. Put it in front of the people who have to be in the room — not the whole mailing list — and let them answer from email.
- [A volunteer shift swap that actually gets a yes](https://relaylink.ai/learn/volunteer-shift-swap): A swap is one shift, one person you are asking, one clock. A list post is how everyone assumes someone else took it, and you still own the desk.
- [A weekly status that actually asks for a decision](https://relaylink.ai/learn/weekly-status-that-asks-for-a-decision): A status email that buries the ask under green dots gets filed. Put the decision in the first sentence, keep the status short, and say what you will do if nobody answers by a dated hour.
- [A youth coach asking a parent — practice time or travel, not both](https://relaylink.ai/learn/youth-coach-and-a-parent): Parents cannot answer a pile. Ask for practice time or for travel, one fact set, one deadline they can meet from a phone in the car line.
- [Acknowledging a RelayLink briefing is not opening it](https://relaylink.ai/learn/acknowledging-is-not-opening): acknowledge_package is the deliberate act a sender can see. Opening a package in the portal, or fetching the magic-link page, is not — scanners prefetch links, and a false "read" would be worse than silence.
- [Add RelayLink to Claude or ChatGPT in about two minutes](https://relaylink.ai/learn/add-relaylink-to-claude-or-chatgpt): RelayLink is a remote MCP server, so there's nothing to install — paste one address into your assistant's connector settings, approve the sign-in, and it gains a correspondence surface.
- [Align teen phone rules across two homes without a fight in text](https://relaylink.ai/learn/teen-phone-rules-across-homes): One house has a charger downstairs at 9. The other does not. Send one proposed rule and the reason a cautious parent needs — not a thread that tries the teen in absentia.
- [An async thesis check-in your advisor can answer](https://relaylink.ai/learn/thesis-advisor-async-check-in): Send chapter status and one decision — continue, cut, or meet — instead of a recap that forces a calendar invite.
- [An HOA board decision that a resident can actually answer](https://relaylink.ai/learn/hoa-board-decision): Board packets bury the call. A briefing puts one choice — paint, reserve, or a contractor — in front of the people who vote, in words a cold reader can follow from email.
- [An open-source user asking a maintainer — without a ninety-comment thread](https://relaylink.ai/learn/open-source-user-and-a-maintainer): Maintainers decide among themselves in public. A user asking one of them needs a scoped question, the work already done, and a way to answer that is not "please read my issue."
- [Ask a classmate to share notes — reciprocal, specific, no guilt](https://relaylink.ai/learn/asking-a-classmate-to-share-notes-fairly): Name the lecture, what you will give back, and a no that costs them nothing. A guilt trip is how you get silence or notes you cannot repay.
- [Ask a godparent for one practical job, with the context they lack](https://relaylink.ai/learn/asking-a-godparent-for-practical-help): They said they wanted to be useful. Send one concrete ask — a Saturday, a ride, a form — and the facts they were never given. Not a vague "be in their life."
- [Ask a grandparent for one opinion — they do not need an app](https://relaylink.ai/learn/asking-a-grandparent-to-weigh-in): They read email. They may have no assistant. Send one question, the two options, and a link — not a novel about the family.
- [Ask a librarian to check a source — claim, citation, question](https://relaylink.ai/learn/asking-a-librarian-to-check-a-source): Send the claim, the citation as you have it, and one question. A librarian can answer from expertise. A folder of PDFs is consulting they did not offer.
- [Ask a sibling to poke holes — you want defects, not agreement](https://relaylink.ai/learn/asking-a-sibling-to-challenge-your-read): You already have a lean. You need the sister who will tell you where it breaks. Ask for a review, label your guesses, and do not reward a polite yes.
- [Ask a TA to review a draft — not for 'thoughts'](https://relaylink.ai/learn/asking-a-ta-to-review-a-draft): A teaching assistant can hunt defects in a section you name. "Thoughts?" asks them to invent the review. Mark inferred claims so they do not grade a sentence you never meant.
- [Ask a teammate to hunt defects in the debate case](https://relaylink.ai/learn/debate-case-review): A case review is a defect hunt, not a pep talk. Tell your teammate which argument to break, cap the time, and keep the model's praise out of the brief.
- [Ask a vendor one scoped question from a student team](https://relaylink.ai/learn/senior-design-vendor-ask): A student team gets a reply when the question is answerable from a spec the vendor already knows. A tour of the project is not that question.
- [Ask an honors committee for a review, not a stamp](https://relaylink.ai/learn/honors-committee-review): A proposal sent for honors should ask the committee to hunt defects. A rubber stamp request hides the work they are supposed to do.
- [Ask an RA or roommate for help — facts, then one ask](https://relaylink.ai/learn/dorm-conflict-ask-for-a-mediator): Separate what happened from how it felt. Then ask one person — an RA or a roommate — for one concrete next step, not a verdict on who is good.
- [Ask for a recommendation letter as a briefing](https://relaylink.ai/learn/asking-for-a-recommendation-letter): A letter packet is a briefing — deadline, what to emphasize, and what you will not invent. Give the writer facts they can use, not a transcript of your anxiety.
- [Ask in-laws for one input — they read email, and that is enough](https://relaylink.ai/learn/asking-in-laws-for-input): They are not in the planning chat. They will answer one question if you respect the hour. Send the question, the two options, and a link — not the whole wedding binder.
- [Ask to swap presentation slots — one concrete trade](https://relaylink.ai/learn/swapping-presentation-slots): Name the two slots, the reason in one fact, and the yes you need. A vague 'can we switch sometime' is how both of you stay on the original day.
- [Asking a beta tester for a verdict, not a vibe](https://relaylink.ai/learn/beta-tester-verdict): They tried the build. You want defects, not encouragement. Name the build, the one flow, and the shape of a review.
- [Asking a colleague to review a grant draft](https://relaylink.ai/learn/grant-review-from-a-colleague): The deadline is real. The draft is long. Ask for defects in the aims or the budget — not a read of every page they will not finish.
- [Asking a designer for a review](https://relaylink.ai/learn/asking-a-designer-for-a-review): A design ask that says take a look produces either silence or a restyle of something you cannot change. Name the kind of review, what is locked, and the one surface you want judged.
- [Asking a former colleague for a warm intro](https://relaylink.ai/learn/former-colleague-warm-intro): A warm intro is still a cold send if they have not accepted you as a contact. Be honest about the channel you actually have, write a note they can forward, and do not pretend you share a workplace.
- [Asking a landlord about a lease change](https://relaylink.ai/learn/tenant-asking-about-a-lease-change): The clause, the date you need an answer, and the one change you want. Not a theory about landlord law. Not a forwarded chat.
- [Asking a landlord about a repair without forwarding the chat](https://relaylink.ai/learn/landlord-repair-without-the-chat): Dates, photos described in words, and one ask — can you send someone this week. Your assistant's theory about the building's plumbing does not belong in that letter.
- [Asking a mentor whether to take the offer](https://relaylink.ai/learn/asking-a-mentor-whether-to-take-the-offer): Two offers, the constraints they cannot see, and an opinion — not a decision they will own. You still pick. They poke the frame.
- [Asking a neighbor about a shared fence — one decision, not a saga](https://relaylink.ai/learn/asking-a-neighbor-about-a-shared-fence): A fence is a line, a material, and who pays. Ask for one of those. Write it so they can say yes or no from email, without reconstructing three summers of chat.
- [Asking a peer to cover on-call](https://relaylink.ai/learn/asking-a-peer-to-cover-on-call): A cover ask that says can you take Friday is easy to ignore and hard to refuse well. Name the window, what is live, what you will do if they say no, and make the no cheap.
- [Asking a PI for lab budget without a meeting](https://relaylink.ai/learn/asking-a-pi-for-lab-budget): The reagent, the amount, the grant line, and a yes or no. A PI who answers email will not open your session about why the project is exciting.
- [Asking a registrar about a hold](https://relaylink.ai/learn/asking-a-registrar-about-a-hold): Your student ID, the hold name as the portal shows it, and one question — what clears it. Not the story of how you got here.
- [Asking a sensitivity reader for a review](https://relaylink.ai/learn/sensitivity-reader-ask): Name the pages, the group you want their eyes on, and that you want defects — not a blessing. Pay them however you already pay them. This letter is the brief.
- [Asking a skip-level to unblock a vendor](https://relaylink.ai/learn/skip-unblock-a-vendor): A vendor is waiting on your company. Your direct path is stuck. A skip-level note that tells the whole saga asks for a meeting. Give them one lever, the cost of silence, and a default you will run.
- [Asking a staff engineer to challenge an estimate](https://relaylink.ai/learn/asking-a-staff-engineer-to-challenge-an-estimate): A review that asks if an estimate looks right invites a polite yes. Ask them to attack the assumptions, list what you already considered, and say which kind of no you want.
- [Asking a union steward a process question](https://relaylink.ai/learn/asking-a-union-steward): What happened, when, who was present, and the one process question. Not a legal theory. Not a vent your assistant polished.
- [Asking an advisor or board member for input between meetings](https://relaylink.ai/learn/asking-an-advisor-or-board-member): Advisors are pattern-matchers with scarce time and stale context, so a vague ask gets an answer that could have been written for any company. Four moves that force a specific reply.
- [Asking an executive for a yes or no](https://relaylink.ai/learn/asking-an-exec-for-a-yes-no): An exec note that asks for thoughts gets a meeting or a silence. Put a binary choice in the first sentence, give the smallest brief that makes a no possible, and name the default if they do not answer.
- [Asking another team to review an RFC](https://relaylink.ai/learn/rfc-review-across-teams): A cross-team RFC fails when the ask is "look at this" and the doc is four thousand words. Name which review you want, list what you already rejected, and make a silent default explicit.
- [Asking facilities for a room with the facts they file](https://relaylink.ai/learn/asking-facilities-for-a-room): Date, time, headcount, setup, and one ask. Facilities does not want your assistant's essay about why the offsite matters.
- [Asking finance to approve a purchase](https://relaylink.ai/learn/finance-approval-for-a-purchase): A purchase ask that arrives as a forwarded quote and a thumbs-up emoji is not a decision. Name what you want to buy, why now, what you already ruled out, and the default if nobody answers.
- [Asking for an academic collaboration](https://relaylink.ai/learn/academic-collaboration-requests): A vague expression of interest asks the recipient to invent the project. Propose the object and the division of labour, name what you bring, and raise credit while it is still cheap.
- [Asking HR a policy question](https://relaylink.ai/learn/asking-hr-a-policy-question): A policy note that asks "is this okay?" forces HR to reconstruct your facts. Send the situation as a short record, then one question the handbook can actually answer — and do not treat the reply as legal advice.
- [Asking IT for access, with the context they actually need](https://relaylink.ai/learn/asking-it-for-access-with-context): An access ticket that says I need prod is a delay you scheduled. Write the system, the reason, the duration, and who already said yes, so a stranger on the queue can decide without paging you.
- [Asking IT for help with the context they replay](https://relaylink.ai/learn/asking-it-helpdesk-with-context): What you clicked, what you saw, what you already tried, and the one thing you need. A ticket that starts with your assistant's theory of the outage goes to the bottom.
- [Asking legal to review a clause](https://relaylink.ai/learn/asking-legal-to-review-a-clause): A note that says looks fine? is not a review request. Name the clause, the risk you care about, and the question you want answered — then wait for the review, not a vibe.
- [Asking your manager for a scope cut](https://relaylink.ai/learn/asking-your-manager-for-a-scope-cut): A scope-cut request that opens with apology and a ticket dump invites a meeting. Lead with what you want dropped, what that protects, and a choice they can make in one reply.
- [Brief the thesis committee before the meeting](https://relaylink.ai/learn/thesis-committee-pre-meeting-brief): Tell them what you want from the hour so it is not a recap. One decision or two review targets, sent in time to read on the train.
- [Build a multi-user AI app with RelayLink](https://relaylink.ai/learn/build-a-multi-user-ai-app-with-relaylink): Authenticate each sender or inbox owner as their own RelayLink account, keep credentials tenant-scoped, and route portal-created webhook envelopes back to the right user.
- [Can AI assistants send emails?](https://relaylink.ai/learn/can-ai-assistants-send-emails): On its own, an assistant produces text — text is not delivered mail. The three ways people wire one up to actually send, what each grants, and the question that matters more than "can it".
- [Can I unsend an AI message? No — and the undo sits before the send](https://relaylink.ai/learn/can-i-unsend-an-ai-message): The honest answer is no. cancel_draft discards a package before it is confirmed, and that is the whole recall story — once a briefing is delivered, a correction on the same thread is the remedy.
- [Can Make Receive RelayLink Webhooks?](https://relaylink.ai/learn/can-make-receive-relaylink-webhooks): Make can pass original JSON and request headers into a scenario, but its documented comparison and response behavior do not establish a complete verification boundary.
- [Can n8n Receive RelayLink Webhooks?](https://relaylink.ai/learn/can-n8n-receive-relaylink-webhooks): n8n can receive RelayLink POSTs, but its documentation does not establish the exact-byte and protected-secret guarantees required for direct verification.
- [Can Zapier Receive RelayLink Webhooks?](https://relaylink.ai/learn/can-zapier-receive-relaylink-webhooks): Zapier exposes an unparsed Catch Raw Hook body and headers, but its automatic response and Code-step secret model make a verifier bridge the safer boundary.
- [Catching up the person who missed standup](https://relaylink.ai/learn/catching-up-the-person-who-missed-stand-up): What changed, what is blocked, and whether they owe a decision. A recap that includes every joke from the call is not a briefing.
- [Changing a neighborhood tool-library rule without a comment pile](https://relaylink.ai/learn/neighborhood-tool-library-rule): A borrowing rule is one sentence with a default. Put the current rule, the proposed rule, and who is harmed if you pick wrong in front of the people who vote — not the whole street.
- [Choosing an elder-care agency: options considered, and who you already ruled out](https://relaylink.ai/learn/choosing-an-elder-care-agency): Three agencies, one already dropped on staffing. Send the remaining two, the reason the third is out, and a single pick — so a sibling is not restarting the search you finished.
- [Claude at work, ChatGPT at home](https://relaylink.ai/learn/claude-at-work-chatgpt-at-home): Use one RelayLink account for a work-to-home self-handoff only when both assistants are authorized, and sanitize or stop when employer data cannot cross.
- [Community garden plot rules that gardeners can vote on](https://relaylink.ai/learn/community-garden-plot-rules): A plot rule is one sentence — water hours, height, or a shared-tool shelf — with a default if nobody answers. Ask the gardeners who hold plots, not the whole neighborhood list.
- [Compare Claude and ChatGPT answers](https://relaylink.ai/learn/compare-claude-and-chatgpt-answers): Compare two assistants against the same question and evidence by separating claims, assumptions, support, and disconfirmers instead of counting agreement or choosing a favorite.
- [Confirming a nanny schedule without a text pile](https://relaylink.ai/learn/nanny-schedule-with-a-parent): Hours, dates, and who is home are facts. "Does Thursday still work?" is the ask. A briefing keeps the chatter out and gives them a way to answer from email.
- [Decision, opinion, or review: choosing what to ask for](https://relaylink.ai/learn/choosing-a-response-shape): A decision transfers ownership of an outcome. An opinion transfers judgment and leaves you holding it. A review hunts defects in something already built. Picking the wrong one is why careful messages get answers that miss.
- [Decisions in open source projects: how to get input without a ninety-comment thread](https://relaylink.ai/learn/decisions-in-open-source-projects): Issue threads are where maintainer decisions go to die — buried framing, self-selected participants, and an exhausted merge at the end. Four moves that fix it, plus the one case where a direct ask beats a thread.
- [Disconnecting an assistant from RelayLink](https://relaylink.ai/learn/disconnecting-a-connected-assistant): /oauth/connections lists the apps you have approved and disconnects one. It covers every approval you gave that app — reconnecting, a second device, a refreshed token — not a single leftover grant.
- [Divide study-group chapters as a decision, not a vibe](https://relaylink.ai/learn/study-group-dividing-chapters): Who reads which chapter is a closed decision with names and dates. Write it down so 'we'll cover it' cannot mean three people skipped the same pages.
- [Documenting a meeting you skipped](https://relaylink.ai/learn/documenting-a-meeting-you-skipped): Catching someone up from notes you were not in the room to hear is inference. Label it that way, separate quotes from conclusions, and do not present a model summary as minutes.
- [Does every RelayLink user need an account?](https://relaylink.ai/learn/does-every-relaylink-user-need-an-account): No. Initiating correspondence or connecting an assistant requires an authenticated account. An unactivated email recipient can receive and reply without signing in.
- [Escalating from success to the account executive](https://relaylink.ai/learn/success-to-ae-escalation): A customer-success note that forwards the ticket history asks an AE to become support. Give them one decision — credit, extend, or hold — plus the three facts that make that decision possible.
- [Exporting and closing a RelayLink account](https://relaylink.ai/learn/export-and-close-a-relaylink-account): You can download your correspondence, then close the account. Close anonymises you and keeps the threads — deleting the row would rewrite the other person's history. Confirmation is your typed address, not a second button.
- [Five situations where an AI briefing beats a long email](https://relaylink.ai/learn/ai-briefing-examples): Worked examples — a cofounder decision, a tax check, a landlord repair, a contractor scope change, and a family trip — showing what a briefing carries and why pasting the chat fails.
- [Follow up with a teacher on what was agreed](https://relaylink.ai/learn/parent-teacher-follow-up): After a conference, write down what was agreed and ask for one next step. The teacher may have no assistant — they can still answer from email.
- [Freelance client communication with AI: how to get a decision, not 'whatever you think is best'](https://relaylink.ai/learn/freelance-client-communication-with-ai): The reply 'whatever you think is best' is not trust — it is a decision nobody owns, and on a freelance engagement the default owner is you. How to hand a client a choice they can actually make.
- [Get a yes on two club budget line items](https://relaylink.ai/learn/club-budget-decision): Two line items and one yes. Close the list, name the total, and ask the treasurer or the board for a decision they can own in a line.
- [Getting a volunteer committee to actually vote](https://relaylink.ai/learn/volunteer-committee-vote): Saturday's meeting produced three versions of the motion and no recorded yes. How to put one vote in front of the people who missed it, with a deadline they can meet from email.
- [Getting answers from your accountant when your AI already has a theory](https://relaylink.ai/learn/getting-answers-from-your-accountant): Your assistant reached a conclusion and now a professional has to check it. How to ask so you get their judgment instead of a rubber stamp on your own framing.
- [Hand a lab notebook to your partner without a file dump](https://relaylink.ai/learn/lab-notebook-handoff-to-a-partner): Say what you did, what is open, and the next action they own. The shared notebook stays where it is. A dump of files is not a handoff.
- [Hand off code context between AI assistants](https://relaylink.ai/learn/hand-off-code-context-between-ai-assistants): Give the next coding assistant a precise checkpoint with repository, branch, commit, relevant paths, test evidence, current failure, constraints, non-goals, and next change.
- [Handing a brief to a teammate who uses a different AI](https://relaylink.ai/learn/handing-a-brief-to-a-teammate-on-another-ai): They work in ChatGPT, you work in Claude, and there is no shared workspace to paste into. A briefing is how the decision, the assumptions, and the open questions leave your session and arrive in theirs.
- [Handing a bug from support to engineering](https://relaylink.ai/learn/support-to-eng-bug-context): A bug note that forwards the ticket asks engineering to become support. Write the outcome the customer wants, the repro, what you already tried, and the one question you need back.
- [Handing off a customer escalation without losing the thread](https://relaylink.ai/learn/customer-escalation-handoffs): An escalation forwarded as a chain forces the new owner to rebuild the story from the bottom up, and the customer to explain it again. Three facts at the top fix most of that.
- [Have one AI review another AI's work](https://relaylink.ai/learn/have-one-ai-review-another-ais-work): Give a second assistant the artifact, requirements, evidence, and a defect-focused ask before sharing the first model's rationale, then make the review decision yourself.
- [Helping an intern choose a project](https://relaylink.ai/learn/intern-project-choice): An intern project note that says pick whatever you like produces a week of stalled work. Give two real options, what each teaches, and a default if they cannot decide by a dated hour.
- [Hire or buy: a family-business decision with a sibling](https://relaylink.ai/learn/family-business-with-a-sibling): The shop needs a second tech or a $12,000 lift. Mark what you know, what you assumed, and ask for one decision — not a recap of who founded what.
- [How a booker asks a venue to hold or release a date](https://relaylink.ai/learn/booker-and-a-venue): A date hold is a yes with a clock. The show, the room, the one technical constraint, and a reply the venue can send from the office inbox.
- [How a bookkeeper asks a client one question that unblocks the monthly close](https://relaylink.ai/learn/bookkeeper-monthly-close-question): Close is a calendar, not a tax theory. One missing fact, the transaction it sits on, and a question the client can answer from their phone.
- [How a consultant sends a recommendation and the options they already ruled out](https://relaylink.ai/learn/consultant-recommendation-to-a-client): The useful package is not the slide of winners. It is the recommendation plus the three options you killed, so the client can own the choice you already did the work on.
- [How a copywriter gets a client to approve voice without a vibe meeting](https://relaylink.ai/learn/copywriter-and-client-voice-approval): Voice approval is a yes on lines, not a feeling about the brand. A short sample, a closed choice, and a note they recognise as yours.
- [How a farm asks a co-op to change a delivery without a phone tag](https://relaylink.ai/learn/farm-and-co-op-delivery-change): A delivery change is a dock decision. One drop, the new window, what happens to the old one, and a yes the co-op can give from the office.
- [How a financial advisor asks a couple for one decision in one mailbox](https://relaylink.ai/learn/financial-advisor-and-a-couple): Two people decide. One briefing still goes to one mailbox. Write the choice so the person who opens it can carry it to the other, not so the software pretends it reached both.
- [How a fractional CFO asks a founder for one cash decision](https://relaylink.ai/learn/fractional-cfo-and-a-founder): A founder will debate the story of the company. The useful ask is one cash choice, the assumption that would flip it, and a date the runway actually cares about.
- [How a game designer asks a publisher to cut or keep one feature](https://relaylink.ai/learn/game-designer-and-publisher): A publisher does not need the whole design again. One feature, what you already ruled out, and a decision that fits the milestone they already named.
- [How a journalist asks a source to confirm one fact without a gotcha](https://relaylink.ai/learn/journalist-source-fact-check): One fact, one confirm, a way to correct it. A check is not a trap, and a briefing is the wrong place to spring a surprise.
- [How a landscaper asks an HOA for one approval a board can actually give](https://relaylink.ai/learn/landscaper-and-an-hoa): A board is a cold reader with a packet habit. One change, the rule you believe you are under, and a yes they can record without a site walk.
- [How a lawyer asks a client for a decision they can actually make](https://relaylink.ai/learn/lawyer-asking-a-client-for-a-decision): A client sitting on a choice does not need another theory. One decision, the facts they can check, and a reply that does not require an account.
- [How a mechanic asks a car owner to choose a repair without a sales pitch](https://relaylink.ai/learn/mechanic-and-car-owner-repair-choice): Observed, judged, and the choice. Two ways to fix what you can see, what each rules out later, and a yes the owner can give from work.
- [How a nonprofit ED asks a board chair for one decision between meetings](https://relaylink.ai/learn/nonprofit-ed-and-board-chair): A chair is a pattern-matcher with stale context. One decision, the two facts that break the default, and an ask that is not a hidden resolution.
- [How a photographer locks a shot list with a client before the day](https://relaylink.ai/learn/photographer-shot-list-with-a-client): A shot list that lives in a thread of 'maybe also' is not a list. Closed must-haves, a few maybes, and one ask the client can answer from email.
- [How a plumber sets planned scope with a homeowner before the van arrives](https://relaylink.ai/learn/plumber-planned-scope-with-homeowner): This is the job you can still define. Options, what you will open, and a yes the homeowner can give from work — not a call from inside the wall.
- [How a realtor gets a buyer to decide on an offer before the deadline](https://relaylink.ai/learn/realtor-and-buyer-offer): An offer deadline is not a mood. One number, the terms that matter, and a written choice the buyer can make without another call.
- [How a recruiter gets a hiring manager to own the next step](https://relaylink.ai/learn/recruiter-and-hiring-manager): A candidate is waiting on a decision the manager has not found in the email. One next step, a closed list, and a deadline that is the candidate's, not yours.
- [How a restaurant partner asks the other owner to change the menu](https://relaylink.ai/learn/restaurant-partner-menu-change): A menu change is a supply decision wearing a taste argument. One item, what it does to the line and the order, and a yes the other partner can own.
- [How a surveyor asks a property owner one question the report cannot answer](https://relaylink.ai/learn/surveyor-and-a-property-owner): The report describes the ground. The owner still has to choose a next step. One finding, in plain words, and a question they can answer without becoming a surveyor.
- [How a translator asks a client to lock one glossary term](https://relaylink.ai/learn/translator-glossary-question): A glossary fight is usually one word wearing a paragraph. The source term, your proposed rendering, and a yes or a correction — not a theory of the whole voice.
- [How a travel planner gets a couple to lock one trip decision](https://relaylink.ai/learn/travel-planner-and-a-couple): Two people, one mailbox, one hold that expires. A closed pair of itineraries and a date beats a brochure nobody answers.
- [How a video producer gets a cut approved without a live review](https://relaylink.ai/learn/video-producer-cut-approval): A cut approval is a yes on a version, not a feeling about the project. Name the cut, what changed, and the one decision that lets you lock picture.
- [How a wedding planner gets a couple to own one vendor decision](https://relaylink.ai/learn/wedding-planner-and-couple-decision): Two people, one mailbox, one vendor choice. A closed pair of options and a date beats a mood board that nobody answers.
- [How an accountant asks a client for the documents that unblock the work](https://relaylink.ai/learn/accountant-asking-a-client-for-documents): The reverse of a client sending a theory. You need paper, named as they would find it, and one list they can finish without a lecture.
- [How an agency PM gets one stakeholder to own the next approval](https://relaylink.ai/learn/agency-pm-and-a-client-stakeholder): A project dies in a thread with three 'looping in's. One person, one approval, a date, and a default if they stay quiet.
- [How an architect gets a material choice from a client without a site meeting](https://relaylink.ai/learn/architect-material-choice-with-a-client): A facade or finish choice dies in a long email with three links. Closed options, what each rules out, and a briefing the client can answer from a phone.
- [How an editor asks an author to accept a cut without starting a fight](https://relaylink.ai/learn/editor-and-author-cut): A cut is a decision, not a verdict on the book. The passage, what it costs the reader, and a bounded choice the author can own.
- [How an electrician explains a permit path so a homeowner can choose](https://relaylink.ai/learn/electrician-permit-path): Two ways the work can run, what each requires from them, and a choice you can record. This is not a reading of the code. It is a written fork before you pull wire.
- [How an interior designer gets a finish choice without another showroom trip](https://relaylink.ai/learn/interior-designer-finish-choice): A finish is a later regret if the ask is a mood. Two options, what each rules out, and the lighting assumption on its own line.
- [How RelayLink sign-in works — an emailed code, no password](https://relaylink.ai/learn/how-relaylink-sign-in-works): RelayLink has no password. You type an address, a code arrives, you enter it. Known and unknown addresses get the same page, so the form cannot be used to ask whether someone has an account.
- [How to ask a busy expert for help](https://relaylink.ai/learn/asking-a-busy-expert-for-help): An expert who owes you nothing has one scarce resource, and it isn't knowledge. Most requests fail by handing over an unframed problem and expecting expertise to supply the framing.
- [How to ask a professor for an extension](https://relaylink.ai/learn/asking-a-professor-for-an-extension): Lead with the facts, one closed ask, and the date you can actually meet. Leave the transcript of your panic session out of the email.
- [How to ask an insurance adjuster one claim question they can answer](https://relaylink.ai/learn/insurance-claim-question-to-an-adjuster): An adjuster is a cold reader with a file number. One fact, one question, and a description of the document — not a packet and not a theory.
- [How to ask for a decision so you actually get one](https://relaylink.ai/learn/how-to-ask-for-a-decision): Most decision requests fail because the reader can't find the decision. Six moves that fix it — and the structure they add up to.
- [How to ask for a technical design review that isn't a rubber stamp](https://relaylink.ai/learn/asking-for-a-technical-design-review): A ninety-second LGTM and a reopened question you settled in week one are the same failure. Four things to state in the request, and where they already have names.
- [How to ask RelayLink for help](https://relaylink.ai/learn/how-to-ask-relaylink-for-help): Open a support request from the public form, your account page, or your assistant. The printed reference is a convenience. The Reply-To token is what lets you add to the thread by email. Acknowledgements echo nothing you typed.
- [How to block and unblock someone on RelayLink](https://relaylink.ai/learn/how-to-block-and-unblock-on-relaylink): Block from the email footer, the contacts page, or your assistant. Nothing is carried either way, and they are not told. Unblock deletes the pair — you are strangers again, not quietly restored as contacts.
- [How to break a cofounder deadlock](https://relaylink.ai/learn/breaking-a-cofounder-deadlock): The third round of the same argument starts from zero because nobody wrote down what the first two settled. Separate the assumptions from the values, and most cofounder deadlocks turn out to be the first kind.
- [How to brief your assistant so the draft lands](https://relaylink.ai/learn/how-to-brief-your-assistant-so-the-draft-lands): The rework that shows up at review usually started earlier in the session. Four habits keep an AI-drafted briefing close enough to right that review stays a check, not a rewrite.
- [How to build an MCP server that's safe to connect](https://relaylink.ai/learn/how-to-build-a-safe-mcp-server): Implementation rules for MCP server authors — narrow tools, honest annotations backed by server-side enforcement, per-call ownership checks, hostile input, neutralized output — worked through against a real seventy-two-tool server.
- [How to check if someone replied — without micromanaging your assistant](https://relaylink.ai/learn/how-to-check-if-someone-replied-without-micromanaging-your-assistant): Two tools answer whether someone replied and what's still open, without rereading a single thread. Neither one skips the review step once there's an actual reply to send.
- [How to connect an agent framework to an MCP server](https://relaylink.ai/learn/how-to-connect-an-agent-framework-to-mcp): What any MCP client has to do — pick a transport, complete the handshake, turn the server's tool list into model-visible tool definitions, and stop tool names colliding across servers.
- [How to connect Claude and ChatGPT to one RelayLink account](https://relaylink.ai/learn/connect-claude-and-chatgpt-to-one-relaylink-account): Connect both assistants as the same RelayLink identity, but give each its own OAuth application connection or named key so either door can be closed without disturbing the other.
- [How to continue a project in a different AI assistant](https://relaylink.ai/learn/continue-a-project-in-a-different-ai-assistant): Continue a project in another AI assistant from a clean checkpoint. Preserve the artifact state, accepted decisions, assumptions, and next task without depending on either provider's chat history.
- [How to correct something your AI got wrong](https://relaylink.ai/learn/how-to-correct-something-your-ai-got-wrong): A message went out with a wrong number, a colder tone than you meant, or a commitment you never intended. What to send next, how fast, and why blaming the assistant makes it worse.
- [How to Deduplicate Webhook Retries](https://relaylink.ai/learn/deduplicate-webhook-retries): Use the stable delivery id as a durable idempotency key, then separate receipt from processing so a crash does not lose work or repeat an external effect.
- [How to divide a group project when nobody can meet](https://relaylink.ai/learn/group-project-when-nobody-can-meet): A 2am group chat is not a plan. Name the pieces, ask each person for one commitment, and put the split where a tired teammate can answer from email.
- [How to follow up without nagging](https://relaylink.ai/learn/how-to-follow-up-without-nagging): A bare bump carries no information and adds only guilt. What a second message has to contain to be worth sending, and why the count matters more than the interval.
- [How to hand off a project so it survives you](https://relaylink.ai/learn/handing-off-a-project): Most handoff docs record where things are and never why they are that way, so the next owner reopens settled decisions and re-breaks fixed things. What to write instead, and which section everyone leaves out.
- [How to handle API keys for AI tools](https://relaylink.ai/learn/api-key-security-for-ai-tools): A key held by an AI client is a password, not a config value. Where it must never go, why one key per integration matters, and the two questions to ask before you create one.
- [How to let your AI assistant contact someone for you — without it speaking as you](https://relaylink.ai/learn/let-your-ai-email-someone): A practical walkthrough of assistant-mediated outreach done safely - drafts you review, provenance the recipient can trust, and a reply path that works over plain email.
- [How to make a mid-renovation decision without deciding on the spot](https://relaylink.ai/learn/deciding-with-a-contractor-on-a-renovation): The crew opened a wall, found something, and wants an answer while you're at work. Here's how to turn that ambush into a short written decision you can still defend in month three.
- [How to move a ChatGPT project to Claude](https://relaylink.ai/learn/move-a-chatgpt-project-to-claude): Move a project from ChatGPT to Claude at a controlled checkpoint rather than attempting an automatic project copy. Inventory artifacts, record decisions and uncertainty, then define the first Claude task.
- [How to move a Claude project to ChatGPT](https://relaylink.ai/learn/move-a-claude-project-to-chatgpt): Move a Claude project to ChatGPT by separating durable artifacts from the working conversation. Package the state, constraints, failed paths, and first assignment without claiming the project itself was copied.
- [How to reply to a RelayLink briefing](https://relaylink.ai/learn/how-to-reply-to-a-relaylink-briefing): Four ways back — email, the package link, the account portal, or your assistant. A block is silent by mail and visible in the browser. An emailed reply still works after the link has expired.
- [How to research in ChatGPT and build in Claude Code](https://relaylink.ai/learn/research-in-chatgpt-build-in-claude-code): Create a concrete technical handoff from ChatGPT research to Claude Code. Define the problem, sourced findings, requirements, non-goals, repository state, relevant paths, and the command used to check completion.
- [How to revoke one AI without disconnecting the others](https://relaylink.ai/learn/revoke-one-ai-without-disconnecting-the-others): First identify whether the assistant uses an OAuth application or a named API key. Close that exact door, verify it stopped, and leave unrelated connections active.
- [How to Rotate a RelayLink Webhook Secret](https://relaylink.ai/learn/rotate-a-relaylink-webhook-secret): A webhook page has a Rotate secret button for the ordinary case: same subscription, same id, same history, new secret, no overlap. This guide also covers the two-subscription overlap strategy for when a zero-message-loss cutover matters more than simplicity.
- [How to route RelayLink webhooks for multiple users](https://relaylink.ai/learn/route-relaylink-webhooks-for-multiple-users): A SaaS receiver can serve many RelayLink accounts, but every account needs its own registered route, subscription secret, and fetch context because the body has no recipient id.
- [How to run a vendor selection without a spreadsheet nobody reads](https://relaylink.ai/learn/running-a-vendor-selection): Weighted scoring matrices mostly launder a decision someone already made. Write the disqualifiers first, ask a small number of questions that actually separate vendors, and keep what you verified apart from what you were told.
- [How to say no in writing](https://relaylink.ai/learn/how-to-say-no-in-writing): A no that comes back for a second round was usually a maybe when it left. Decide before you write, lead with the answer, give one real reason, and keep the warmth separate from the refusal.
- [How to secure a RelayLink webhook receiver](https://relaylink.ai/learn/secure-a-webhook-receiver): Build the receiver as a narrow notification boundary: authenticate every POST, reject stale requests, deduplicate retries, fetch separately, and log no secrets.
- [How to set a deadline someone will believe](https://relaylink.ai/learn/how-to-set-an-honest-deadline): A date is only half a deadline. The half that makes it credible — what you will do when it passes — is the half almost everyone leaves out.
- [How to Set Up RelayLink Webhooks](https://relaylink.ai/learn/how-to-set-up-relaylink-webhooks): Set up a public HTTPS receiver, register it from the RelayLink account page, store the one-time signing secret, and prove the path with a test event.
- [How to share an AI conversation with someone (and when you shouldn't)](https://relaylink.ai/learn/how-to-share-an-ai-conversation): Share links and copy-paste both hand over the whole session — usually more than you meant to share and less than the other person needs. Here are the options, and the one built for getting an answer back.
- [How to share context between ChatGPT and Claude](https://relaylink.ai/learn/share-context-between-chatgpt-and-claude): ChatGPT and Claude cannot literally share a session or hidden context. You can give the second assistant a reviewed briefing that states the facts, assumptions, decisions, and question it needs.
- [How to state an assumption so someone can correct it](https://relaylink.ai/learn/how-to-state-an-assumption): The load-bearing premise in most messages is a subordinate clause nobody notices. How to surface it, attribute it, and make it cheap to correct.
- [How to stop AI-generated spam when it knows your name](https://relaylink.ai/learn/how-to-stop-ai-generated-spam): Personalized machine outreach defeats the filters and the instincts you built for the old kind. What still works for a recipient — and why the durable fix has to sit at the sending end.
- [How to switch AI tools without copying the whole chat](https://relaylink.ai/learn/switch-ai-tools-without-copying-the-whole-chat): Switch between ChatGPT and Claude with a minimal, reviewed handoff instead of exposing the entire working conversation. Keep the facts, decisions, uncertainty, and next ask; leave unrelated exploration behind.
- [How to tell if a cold briefing is worth your time](https://relaylink.ai/learn/how-to-tell-if-a-cold-briefing-is-worth-your-time): Not every briefing arrives from someone you know. A checklist for judging a cold one before you spend time on it, and why RelayLink's caps mean it already cost the sender something to reach you.
- [How to test a RelayLink webhook endpoint](https://relaylink.ai/learn/test-a-relaylink-webhook-endpoint): Send a signed webhook.test event from the account portal, inspect the exact body and headers, and prove your receiver is ready before a package depends on it.
- [How to transfer context from ChatGPT to Claude](https://relaylink.ai/learn/transfer-context-from-chatgpt-to-claude): Move the useful result of a ChatGPT research or analysis session into Claude as a reviewed decision brief. Preserve sources, assumptions, uncertainty, and the exact question Claude should answer.
- [How to transfer context from Claude to ChatGPT](https://relaylink.ai/learn/transfer-context-from-claude-to-chatgpt): Hand a Claude coding or long-form session to ChatGPT without pasting the working conversation. Capture the current state, fixed constraints, rejected paths, artifacts, and one clear question.
- [How to Troubleshoot RelayLink Webhooks](https://relaylink.ai/learn/troubleshoot-relaylink-webhooks): Trace a missing RelayLink webhook from the account-page test through HTTPS reachability, signature handling, response status, retries, and automatic disablement.
- [How to use ChatGPT for research and Claude for coding](https://relaylink.ai/learn/use-chatgpt-for-research-and-claude-for-coding): Turn a ChatGPT research session into a coding brief for Claude without presenting model output as verified fact. Carry sources, uncertainty, requirements, non-goals, repository references, and a test target.
- [How to use Claude and ChatGPT together](https://relaylink.ai/learn/how-to-use-claude-and-chatgpt-together): Use Claude and ChatGPT as separate working environments with deliberate roles and reviewed handoffs. One RelayLink account gives both assistants the same correspondence inbox without pretending they share a mind.
- [How to Verify RelayLink Webhook Signatures](https://relaylink.ai/learn/verify-relaylink-webhook-signatures): Verify the exact raw body with HMAC-SHA256, compare the MAC in constant time, enforce your timestamp policy, and deduplicate the signed delivery id.
- [How to write a note your AI can't improve](https://relaylink.ai/learn/how-to-write-a-note-your-ai-cant-improve): The sender's note is the one field in a briefing built for your own words, not your assistant's. Practical advice for keeping it that way — and for when sending the AI's version is the right call.
- [How to write a subject line that isn't a lie](https://relaylink.ai/learn/how-to-write-a-subject-line-that-isnt-a-lie): Subject lines drifted into advertising, so readers learned to discount them. The fix is to make them descriptive again — the topic, the ask type, and the distinguishing word first.
- [How to write a TL;DR people actually rely on](https://relaylink.ai/learn/how-to-write-a-tldr): A TL;DR is a verdict, not a topic label, and some readers will act on it without opening anything else. The rules that make it safe for them to do that.
- [How to write for someone with zero context](https://relaylink.ai/learn/how-to-write-for-a-cold-reader): You cannot un-know your own background, so you cannot feel what is missing from your draft. Four mechanical tests that substitute for the perception you do not have.
- [IEP notes to the other parent: what the meeting said, not what you concluded](https://relaylink.ai/learn/iep-notes-to-the-other-parent): You were in the room. They were not. Send the team's sentences and the one next step — describe the document, do not attach the PDF.
- [Inheritance admin with a sibling: facts first, no legal advice](https://relaylink.ai/learn/inheritance-admin-with-a-sibling): Someone has to call the bank and someone has to find the deed. Send the checklist and what the paper actually said — not your assistant's reading of a will summary.
- [Inviting someone to be a RelayLink contact](https://relaylink.ai/learn/inviting-someone-to-be-a-contact): A contact request is how two account holders open a channel. Ask from the assistant or the contacts page. They accept — you cannot accept your own request. The request carries no note. Asking twice does not send a second email.
- [Is it safe to let AI write your emails?](https://relaylink.ai/learn/is-it-safe-to-let-ai-write-emails): Safe is four separate questions — accuracy, attribution, access, autonomy — with four different answers. The mitigation for each, including the one no product solves.
- [Kids' activities across two homes: who signs, who pays, who drives](https://relaylink.ai/learn/kids-activities-across-two-homes): Signup closes Friday. One house has the form, the other has Saturday. Send the deadline, the cost, and the two jobs — not a thread about who is the fun parent.
- [Kill, hold, or run — one editor, one story](https://relaylink.ai/learn/student-newspaper-edit-decision): A story on deadline needs a closed choice from one editor. Kill, hold, or run. A group chat of vibes is how the piece misses the issue.
- [Make your case for a lineup change — their call](https://relaylink.ai/learn/asking-a-coach-about-a-lineup-change): Send the case you can defend — minutes, matchups, what you will do either way. Ask for a decision the coach owns. Do not ask them to soothe you.
- [Making a decision for someone who is on leave](https://relaylink.ai/learn/decision-for-the-person-on-leave): Leave is not a pause button on every choice. Write what you decided, why, what would reopen it, and what you deliberately left untouched, so they can return to a record instead of a mess.
- [Making decisions across time zones](https://relaylink.ai/learn/cross-timezone-team-decisions): A decision spread across fifteen hours of time zones costs a day per round trip. How to write one so it needs zero round trips, and what to do about the people who are asleep when it lands.
- [Named API keys on RelayLink — one credential per agent](https://relaylink.ai/learn/named-api-keys): Keys are named rows on your account. Issue one in the browser, revoke one without touching the others. Issuing is not an assistant tool, because the secret exists in the clear exactly once.
- [OAuth vs API key for RelayLink](https://relaylink.ai/learn/oauth-vs-api-key-for-relaylink): Use RelayLink OAuth when a client supports browser sign-in and consent. Use a named API key for an MCP client that can attach a header but cannot complete that flow.
- [Office hours you cannot attend — one written question](https://relaylink.ai/learn/office-hours-async-when-you-cannot-attend): If you cannot sit in the room, send one question a lecturer can answer in writing. A diary of the week is not a substitute for the queue.
- [One RelayLink account for multiple AI assistants](https://relaylink.ai/learn/one-relaylink-account-multiple-ai-assistants): A RelayLink account represents one person and email identity, not one assistant. Connect several assistants to share correspondence while keeping a separate credential for each door.
- [One RelayLink account with multiple webhook endpoints](https://relaylink.ai/learn/one-account-multiple-webhook-endpoints): Register several named receivers for one account when separate systems need the arrival signal. Each endpoint has its own secret, delivery history, controls, and failure state.
- [Pet care while you travel: instructions plus one confirmation](https://relaylink.ai/learn/pet-care-when-you-travel): Someone is watching the dog. Send the meds, the vet, the alarm code — and ask for one yes that they can do those days. Not a novella, not a group chat.
- [Pick a reunion date and place when the family is scattered](https://relaylink.ai/learn/family-reunion-logistics): Four households, two candidate weekends, one cabin hold. Send the comparison and ask the person whose calendar actually blocks the date — not a twenty-person poll.
- [Picking a contact in Claude Code](https://relaylink.ai/learn/picking-a-contact-in-claude-code): Claude Code lists RelayLink contacts under @ and offers send-to as a slash command. Both resolve the person on the server and stop at a draft. Neither is a send. The list is yours — not everyone else's correspondents.
- [Picking an alumni reunion date without a thirty-comment poll](https://relaylink.ai/learn/alumni-reunion-date): A reunion date is two weekends, not a blank calendar. Put both on the page with the one constraint that matters — a hall hold, a conflict, a travel window — and ask the people who must book.
- [Pickup swaps without turning the calendar into a group chat](https://relaylink.ai/learn/two-household-calendars): Two houses, one week, three conflicts. Send the dated swap and the one ask — not a thread that relitigates every Tuesday.
- [Plan a family trip without a twenty-person group thread](https://relaylink.ai/learn/family-trip-without-a-group-thread): Dates, a budget cap, and the destination you already dropped. Send one decision — which week — instead of another chat that forgets why you ruled out the beach.
- [Plan in ChatGPT, implement in Claude](https://relaylink.ai/learn/plan-in-chatgpt-implement-in-claude): Turn a plan made in ChatGPT into a bounded implementation brief for Claude, with the decisions, repository references, tests, and next coding ask made explicit.
- [Podcast guest prep that a guest can answer once](https://relaylink.ai/learn/podcast-guest-prep): A guest should not reconstruct your show from a thread. Send the date, length, topics you will stay on, and one thing you need back — bio line, release, or a no.
- [Preserve assumptions when switching AI assistants](https://relaylink.ai/learn/preserve-assumptions-when-switching-ai-assistants): Carry stated and inferred assumptions with their origin, impact, disconfirmers, and recheck step so another assistant does not mistake a working premise for a fact.
- [Preserve decisions when switching AI assistants](https://relaylink.ai/learn/preserve-decisions-when-switching-ai-assistants): Carry the decision, rejected options, owner, consequences, and reopen condition so another assistant can continue the work without treating settled choices as forgotten ideas.
- [Preserve sources when switching AI assistants](https://relaylink.ai/learn/preserve-sources-when-switching-ai-assistants): Keep a source ledger with identity, URL, quoted claim, check status, and uncertainty so the next assistant can distinguish evidence from the first model's summary.
- [Production and sandbox RelayLink in one assistant](https://relaylink.ai/learn/production-and-sandbox-relaylink-in-one-assistant): Connect production and sandbox as separate MCP hosts, use their distinct handles for intent, and verify the account and server with whoami before drafting.
- [Professional outreach that isn't spam](https://relaylink.ai/learn/networking-outreach-that-isnt-spam): Outreach worth sending doesn't scale, and that is the mechanism rather than the flaw. Five moves that don't survive being mass-sent, plus what a first contact actually costs the person sending it.
- [RelayLink for agencies with many clients](https://relaylink.ai/learn/relaylink-for-agencies-with-many-clients): Agency staff use their own RelayLink identities across many client relationships. Clients may be accepted contacts or email-only recipients, without shared-account impersonation.
- [RelayLink for consultants with multiple clients](https://relaylink.ai/learn/relaylink-for-consultants-with-multiple-clients): One consultant account can manage many client contacts and threads. Clients can stay on email, while the consultant approves each briefing before it is delivered.
- [RelayLink for internal company AI assistants](https://relaylink.ai/learn/relaylink-for-internal-company-ai-assistants): RelayLink can carry employee-owned briefings across teams when each person keeps their own identity and approves what crosses. It does not replace live team discussion.
- [RelayLink for SaaS AI assistants](https://relaylink.ai/learn/relaylink-for-saas-ai-assistants): RelayLink fits SaaS assistants that help an identified person prepare and approve correspondence to another person. It is not an autonomous or bulk messaging API.
- [RelayLink nicknames are private — including inside the package](https://relaylink.ai/learn/nicknames-are-private): The word you call a contact lives on your side of the pair only. It drives how you address them. It does not appear to them unless you say it can — and the server refuses a draft that puts it in the briefing they would read.
- [RelayLink outbound webhooks: the envelope, never the briefing](https://relaylink.ai/learn/outbound-webhooks): Register a URL on your account and RelayLink POSTs a signed envelope when a package arrives for you — ids, sender, topic and time, never the note or the TL;DR. Unrelated to inbound email webhooks.
- [RelayLink vs a Shared Slack Channel with a Client](https://relaylink.ai/learn/relaylink-vs-shared-slack-channel-with-a-client): Choose a shared Slack channel for live group coordination and RelayLink for direct, approved correspondence with a stable ask, consent boundary, and provenance.
- [RelayLink vs. giving your AI agent its own email account](https://relaylink.ai/learn/relaylink-vs-diy-agent-email): The fastest way to let an agent send email is to hand it a mailbox. Sometimes that is exactly right. A comparison of what the DIY route wins and what it quietly costs.
- [RelayLink vs. sharing a ChatGPT link: when each one is the right call](https://relaylink.ai/learn/relaylink-vs-sharing-a-chatgpt-link): A shared ChatGPT link takes one click and shows exactly what was said. A RelayLink briefing takes more effort and asks a specific question instead. Here's the honest tradeoff, and when each one earns its use.
- [RelayLink Webhook Failure Runbook](https://relaylink.ai/learn/relaylink-webhook-failure-runbook): Diagnose a silent RelayLink receiver from the account-page test through HTTP status, exact-byte HMAC, timeout, retry, automatic disablement, and recovery.
- [RelayLink Webhook for Microsoft Teams Notifications](https://relaylink.ai/learn/relaylink-webhook-for-microsoft-teams-notifications): Use a verified bridge and a Teams Workflow to announce RelayLink arrivals without exposing the briefing or treating a channel post as a read.
- [RelayLink Webhook for Slack Notifications](https://relaylink.ai/learn/relaylink-webhook-for-slack-notifications): Bridge RelayLink arrivals into a Slack channel as fixed, escaped notifications without copying the briefing or creating an unapproved reply path.
- [RelayLink Webhook to CrewAI Flow](https://relaylink.ai/learn/relaylink-webhook-to-crewai-flow): Wake a CrewAI Flow from a RelayLink arrival without turning webhook metadata into agent instructions or losing the event during a restart.
- [RelayLink Webhook to LangGraph Agent](https://relaylink.ai/learn/relaylink-webhook-to-langgraph-agent): Use a signed RelayLink webhook to wake a durable LangGraph run, then fetch the package through MCP and keep outside correspondence out of system instructions.
- [RelayLink Webhook with a Temporal Workflow](https://relaylink.ai/learn/relaylink-webhook-with-temporal-workflow): Turn a signed RelayLink arrival into a durable Temporal Workflow or Signal while preserving transport deduplication, account-scoped fetching, and human approval.
- [RelayLink Webhook with Inngest](https://relaylink.ai/learn/relaylink-webhook-with-inngest): Verify a RelayLink callback in your own receiver, then emit a minimal idempotent Inngest event that fetches the package through MCP in durable steps.
- [RelayLink Webhooks on AWS Lambda](https://relaylink.ai/learn/relaylink-webhooks-on-aws-lambda): Receive RelayLink through an API Gateway Lambda proxy integration, decode its body exactly once, verify HMAC, and claim delivery before returning success.
- [RelayLink Webhooks on Azure Functions](https://relaylink.ai/learn/relaylink-webhooks-on-azure-functions): Use an Azure Functions Node.js HTTP trigger to read one ArrayBuffer, verify RelayLink before JSON parsing, and persist an idempotent handoff.
- [RelayLink Webhooks on Cloudflare Workers](https://relaylink.ai/learn/relaylink-webhooks-on-cloudflare-workers): Verify RelayLink at the edge with Request.arrayBuffer(), Web Crypto HMAC verification, a Worker secret binding, and durable idempotent handoff.
- [RelayLink Webhooks on Google Cloud Run](https://relaylink.ai/learn/relaylink-webhooks-on-google-cloud-run): Deploy a public Cloud Run HTTP function, verify RelayLink from req.rawBody, inject the signing secret from Secret Manager, and queue work durably.
- [RelayLink Webhooks with Express](https://relaylink.ai/learn/relaylink-webhooks-with-express): Mount express.raw before JSON middleware, verify RelayLink against the Buffer with Node.js crypto, and commit an idempotent inbox record.
- [RelayLink Webhooks with FastAPI](https://relaylink.ai/learn/relaylink-webhooks-with-fastapi): Read FastAPI's Request body as bytes, verify RelayLink with hmac.compare_digest, then atomically claim and queue each delivery.
- [RelayLink Webhooks with Next.js Route Handlers](https://relaylink.ai/learn/relaylink-webhooks-with-nextjs-route-handlers): Build a POST Route Handler that reads exact request bytes in the Node.js runtime, verifies RelayLink HMAC, and queues work outside the request.
- [Research with ChatGPT, write with Claude](https://relaylink.ai/learn/research-with-chatgpt-write-with-claude): Move from research to drafting with an evidence brief that preserves source details, quoted claims, uncertainty, audience, and the writing job without forwarding the research chat.
- [Review a RelayLink draft in the portal](https://relaylink.ai/learn/review-a-draft-in-the-portal): When you edit the note on /account/drafts/{id}, you are signed in with an emailed code — no assistant holds that session. Leave the draft alone and it is approved. Change it and it is verbatim. Emptying the box is refused.
- [Run a classroom peer review that is actually a review](https://relaylink.ai/learn/classroom-group-peer-review): Ask a classmate to hunt defects in a named section, with a time cap and a date. 'Nice job' is not a peer review, and the course already knew that.
- [Run a student-government vote without a midnight meeting](https://relaylink.ai/learn/student-government-vote-async): Put the motion, the options, and the deadline in writing. Members can reply from email — they do not need an account to cast a yes or no.
- [Scout permission from a parent, in writing they can answer](https://relaylink.ai/learn/scout-permission-from-a-parent): A permission ask is one trip, one date, one yes. Put the facts a parent needs in the note — not a packet they cannot open on a phone — and let them reply from email.
- [Selling the childhood home: siblings, a list price, one decision](https://relaylink.ai/learn/selling-the-childhood-home): The agent wants a number by Friday. Send the comps you trust, the number you lean toward, and a closed choice — not a tour of every memory in the hallway.
- [Send a RelayLink briefing to yourself](https://relaylink.ai/learn/send-a-briefing-to-yourself): One RelayLink account can hold Claude at work and ChatGPT at home. Address a package to your own email and it lands as unread inbound until any of your assistants pulls it — no second account, no contact request, no email.
- [Sending a hiring-loop debrief to your partner](https://relaylink.ai/learn/hiring-loop-debrief-to-a-partner): A debrief that reads like a story forces the hiring partner to guess what you actually recommend. Separate what the candidate said from what you inferred, and close with a hire / more loop / no.
- [Set tutoring scope with a parent — hours, subject, done](https://relaylink.ai/learn/tutoring-scope-with-a-parent): Agree hours, the subject, and what 'done' means before the first session. A parent can confirm from email. A vague 'help with school' has no end.
- [Setting team norms for AI-drafted communication](https://relaylink.ai/learn/setting-team-norms-for-ai-communication): A team that never discusses this doesn't end up with no policy — it ends up with one invisible policy per person. Five norms worth writing down, and the honest reason most norm lists rot within a quarter.
- [Share a report-card read: the grades as stated, your conclusion as yours](https://relaylink.ai/learn/sharing-a-report-card-read): The card is in. Send the marks and one sentence of your read — not a file, not a verdict the other parent never got to form.
- [Share what the clinician said, not what your assistant concluded](https://relaylink.ai/learn/sharing-a-medical-update-with-family): A sibling will decide the next step. Send the clinician's words as stated, and mark every inference so they can tell the difference.
- [Should a team share one RelayLink account?](https://relaylink.ai/learn/should-a-team-share-one-relaylink-account): Generally, no. One account presents one person's email and display identity. Several assistants for that person are fine; several people should keep distinct accounts.
- [Split a family expense: send the numbers as facts, the split as the ask](https://relaylink.ai/learn/splitting-a-family-expense): The roof invoice is in. The question is who pays what. Put the figures in the stated column and ask for the split — not a debate about who uses the house more.
- [Swapping a shift without the group chat](https://relaylink.ai/learn/shift-swap-without-the-group-chat): One person, one shift, one yes. A group thread is where asks go to die. A briefing is a question somebody can answer from a break.
- [Tell family you are moving: FYI, or ask for help — not both at once](https://relaylink.ai/learn/telling-family-you-are-moving): The lease is signed. Decide whether this message is an announcement or a request for Saturday hands. Mixing them is how nobody books a truck and everyone sends a heart.
- [Tell the other parent the schedule changed — dated, neutral, one ask](https://relaylink.ai/learn/telling-the-other-parent-about-a-schedule-change): The Tuesday pickup moved. Send the date, the reason in one line, and the one thing you need back. Keep your private name for them off the page.
- [Tell your lab partner what changed in the protocol](https://relaylink.ai/learn/lab-partner-protocol-change): A method change is a decision, not a chat dump. Say what you altered, why, and the one thing you need them to confirm before the next run.
- [Telling a client about a scope change without starting an argument about blame](https://relaylink.ai/learn/telling-a-client-about-a-scope-change): The scope message usually gets written defensively at the end of a long day — apology first, number buried. Lead with the change and its cost, split the history from the decision, and give a bounded choice.
- [Telling a peer you will miss a deadline](https://relaylink.ai/learn/telling-a-peer-you-will-miss-a-deadline): A late note to a peer that opens with apology and history asks them to manage your feelings. Lead with the new date, what you still owe them, and the one thing you need from them now.
- [Telling a vendor PM that the integration slipped](https://relaylink.ai/learn/vendor-pm-integration-slip): A slip note to a vendor that hides the new date behind a technical story asks them to guess what they can tell their side. Lead with the date, what you still need from them, and what they can stop waiting on.
- [Telling sales that engineering slipped](https://relaylink.ai/learn/telling-sales-engineering-slipped): A slip note that dumps the ticket board asks sales to become project managers. Lead with what they can still promise, what they must stop promising, and the one decision you need from them.
- [The /docs page is generated from the running server](https://relaylink.ai/learn/the-docs-page-is-generated): /docs is not a second copy of the tools. It reads the server — the catalogue, the instructions sent at initialize, the provenance labels, the five send refusals. /llms-full.txt is the same facts as markdown. A handwritten list on the landing page is what this exists to stop.
- [The on-call handoff as a briefing](https://relaylink.ai/learn/on-call-handoff-as-a-briefing): A rotating pager fails when the incoming person inherits a chat scroll. Write what is on fire, what is only being watched, what was already tried, and the one decision that is still open.
- [The package page a recipient opens from email](https://relaylink.ai/learn/the-package-link-recipients-see): /p/{token} is the first thing a stranger sees. It is a page, not an app — no script, every field encoded, an expired link looks the same as one that never existed. Replying to the email still works after the page does not.
- [The RelayLink shorthand opens a draft, never a send](https://relaylink.ai/learn/the-at-relaylink-shorthand): No chat client lets an MCP server register a mention. The shorthand works because server instructions teach it — and those instructions say it drafts for your approval, the same as any other compose.
- [The RelayLink welcome email](https://relaylink.ai/learn/the-welcome-email): The first time an account becomes real, one mail says what to do next. It goes out once, on sign-in or first OAuth consent — not when an operator provisions you, and not when you issue a key. Four steps and four links, all RelayLink's.
- [The standup that should have been a briefing](https://relaylink.ai/learn/standup-that-should-have-been-a-briefing): A fifteen-minute standup used to transfer status is a tax on eight calendars. Write the status once, put the one decision at the top, and keep the call for the disagreement writing could not close.
- [They say they never got it](https://relaylink.ai/learn/they-say-they-never-got-it): The causes of a missing message, in the order worth checking. Most are boring, one of them is usually yours, and asserting delivery before you have looked is the expensive mistake.
- [Troubleshooting an MCP connection that won't work](https://relaylink.ai/learn/troubleshooting-mcp-connection-problems): A bottom-up debugging ladder for remote MCP servers — reachability, transport, the auth header, stale tool lists, logs — plus the RelayLink symptoms that look like failures and aren't.
- [Two authors splitting chapters without a call](https://relaylink.ai/learn/two-authors-splitting-chapters): Who writes what, by when, and what "done" means. A chapter plan is a decision record, not a vibe about the book.
- [Two camps, one week, one decision between parents](https://relaylink.ai/learn/summer-camp-choice-between-parents): Both camps want a deposit for the same week. Send the two programs, the one constraint that actually matters, and a closed choice — not a thread about whose childhood was better.
- [Two coaches, one player decision](https://relaylink.ai/learn/two-coaches-player-decision): Playing time, a position change, or who travels. Facts you both saw. One decision. Keep the parent chat out of it.
- [Two founders, two assistants, no shared workspace](https://relaylink.ai/learn/two-founders-on-different-ais): You live in Claude. They live in ChatGPT. There is no shared Slack and no shared project. A briefing is how a hiring plan or a pricing call travels anyway.
- [Two fundraisers, one PTA vote](https://relaylink.ai/learn/pta-fundraiser-choice): Put two fundraiser options on one ballot with a deadline. Members can vote from email — they do not need an account or an evening meeting.
- [Two homes, one school choice: send the options, not the fight](https://relaylink.ai/learn/coparenting-a-school-decision): The teacher named two programs. The other parent needs those options and what was actually said — not the argument you had in the car.
- [Two house offers, one deadline, and a partner who is in the air](https://relaylink.ai/learn/house-offers-while-a-partner-travels): Two written offers, the math, and an expiration. Send the decision they can answer from a hotel wifi — after you have approved the exact package.
- [Two internship offers — ask a mentor what they see](https://relaylink.ai/learn/internship-offer-with-a-mentor): Send two offers and ask which risks they see. That is an opinion. You still decide. Do not ask them to pick your life in a line.
- [Two study-abroad housing options — decide with someone at home](https://relaylink.ai/learn/study-abroad-housing-choice): Put two housing options, costs, and dates in writing so a parent or partner can choose with you. They can answer from email; they do not need your campus login.
- [Two wedding quotes, one date, one decision with your fiancé](https://relaylink.ai/learn/wedding-vendor-with-your-fiance): Two caterers, a hold that expires Friday, and a partner who has not seen the menus. Send the comparison and the one pick — you still approve what leaves.
- [Unsent drafts on RelayLink](https://relaylink.ai/learn/unsent-drafts-on-relaylink): draft_package writes a draft. It is not sent until you approve it — in the conversation, or on the Approvals page. An unconfirmed draft expires after twenty-four hours. cancel_draft throws it away sooner.
- [Use one AI for coding and another for documentation](https://relaylink.ai/learn/use-one-ai-for-coding-and-another-for-documentation): Give a documentation assistant a reviewed code-state brief with the exact version, observed behavior, intended audience, safe examples, limits, and verification commands kept together.
- [Use one AI for debugging and another for review](https://relaylink.ai/learn/use-one-ai-for-debugging-and-another-for-review): Let one assistant reduce a failure to reproducible evidence, then give another assistant a neutral review brief that asks for defects without preloading the first model's argument.
- [Use one RelayLink API key per AI assistant](https://relaylink.ai/learn/one-api-key-per-ai-assistant): Give every key-based assistant a named credential of its own. The account stays shared, while each key's name, last-used signal, and revocation remain independent.
- [Using the RelayLink account portal without an assistant](https://relaylink.ai/learn/using-the-account-portal): Your inbox, threads, contacts, compose form and keys live at /account. Sign in with an emailed code — no assistant required — and every send still runs the same consent gate as MCP.
- [Webhook vs Email Notification](https://relaylink.ai/learn/webhook-vs-email-notification): A webhook triggers a machine workflow; an email alerts a person and gives them a reading and reply path. RelayLink can send both without treating either delivery as a read.
- [Webhook-Triggered AI Agent Safety Checklist](https://relaylink.ai/learn/webhook-triggered-ai-agent-safety-checklist): A production checklist for agents awakened by webhooks: authenticate bytes, deduplicate durably, fetch with least privilege, isolate untrusted content, and require approval.
- [Webhooks vs Polling for AI Agents](https://relaylink.ai/learn/webhooks-vs-polling-for-ai-agents): Use webhooks to trigger a self-hosted agent when a package arrives, polling to ask what is waiting, or both. Either path still fetches package content through MCP.
- [What a partner needs to know about coparenting logistics](https://relaylink.ai/learn/introducing-coparenting-logistics-to-a-new-partner): They are entering a calendar that was built without them. Send the pickup map and the one ask — not the whole history of the other household.
- [What a RelayLink notification email actually contains](https://relaylink.ai/learn/what-the-notification-email-contains): The mail that arrives when someone sends you a briefing carries the note, the ask, the summary, the message and a link — every part labelled with who wrote it — and differs if you already have an account. New recipients are told what RelayLink is before they read anything; account holders get Open in Claude and Open in ChatGPT with a fixed prompt.
- [What a RelayLink thread is](https://relaylink.ai/learn/what-a-relaylink-thread-is): A thread is one conversation, many briefings. list_threads lists them. thread_status says who is waiting. A hand-off to yourself has one participant and is awaiting you until an assistant pulls it.
- [What an unsubscribe link actually signals](https://relaylink.ai/learn/unsubscribe-links-and-what-they-signal): An unsubscribe link is a promise about the sender's future behaviour. Whether it is worth anything depends entirely on where the promise is enforced — and most are enforced by the sender.
- [What FYI means on a RelayLink thread](https://relaylink.ai/learn/what-fyi-means-on-relaylink): An FYI is a briefing that does not expect a reply. Once it has been seen — pulled, acknowledged, or replied anyway — the thread is idle, not awaiting anyone. A hand-off to yourself is treated as an FYI whatever shape you gave it.
- [What get_package does — and what it does not](https://relaylink.ai/learn/what-get-package-does): get_package is how an assistant reads a briefing. It records that their assistant fetched it, which is what thread_status uses for "have they seen it." Opening the portal page is a different act and does not write that line.
- [What happens when someone blocks you](https://relaylink.ai/learn/what-happens-when-someone-blocks-you): Nothing tells you at the time. The next send fails, naming the block. They can lift it; that deletes the pair rather than putting you back in their contacts. Until then, nothing is carried either way.
- [What list_contacts actually shows (and why most correspondence never needs it)](https://relaylink.ai/learn/setting-up-your-first-contact-pair): A contact pair is the requirement for standing correspondence between two RelayLink accounts, and it takes two deliberate acts — somebody asks, somebody accepts. Here's what actually determines whether someone shows up in list_contacts, why correspondence alone never will, and why it usually doesn't matter.
- [What RelayLink carries, and what it does not](https://relaylink.ai/learn/attachments-are-out-of-scope): A briefing can now carry a plain-text or Markdown document, checked before it is stored and read only when your assistant asks. A PDF, a deck or a spreadsheet still lives where it already lives. A reply that is only an email attachment still gets a courtesy notice, not a silent drop.
- [What to do when you receive an AI briefing](https://relaylink.ai/learn/what-to-do-when-you-receive-an-ai-briefing): An AI briefing lands in your inbox as a normal email — a note, a TL;DR, and a link, nothing to install. Here's what to check before you answer, and how to reply either way.
- [What you see after you send a briefing](https://relaylink.ai/learn/what-the-sender-sees-after-a-send): The package page lists sent, whether the notification was accepted or bounced, whether their assistant pulled it, acknowledged, replied. A fetched link is never on that list — scanners open URLs before people do.
- [What's new, the RelayLink inbox, and why every tool mentions unread mail](https://relaylink.ai/learn/whats-new-and-the-inbox): Hosted assistants cannot be pushed a notification. whats_new and check_inbox are how you learn what arrived; every other tool appends an unread count so news travels on a call you were already making.
- [When a contact request goes unanswered](https://relaylink.ai/learn/when-a-contact-request-goes-unanswered): They are not notified that they ignored you. Asking again does not send a second email. You find out the next time you try to send, or when whats_new on their side finally gets an accept. Withdraw if you want the pending pair gone.
- [When not to use AI to write a message](https://relaylink.ai/learn/when-not-to-use-ai-for-a-message): Five kinds of message where drafting with an assistant makes the message worse, not faster — and what is left over, which is where drafting genuinely earns its keep.
- [Who is on the paper — ask it as a decision](https://relaylink.ai/learn/research-group-authorship): Authorship is a decision with labelled assumptions, not a hallway vibe. Write the proposed list, mark what you do not know, and ask the group to accept, edit, or meet. This is not legal advice.
- [Who you are connected as on RelayLink](https://relaylink.ai/learn/who-you-are-connected-as): whoami answers which account and which RelayLink host this connection is. set_my_name is what recipients see. A placeholder name is recorded on the account — not guessed from the shape of your address.
- [Why links and images in a briefing arrive defused](https://relaylink.ai/learn/why-links-and-images-arrive-defused): Relayed text is someone else's words. Images that would fetch a URL are removed. Markdown links become readable text. Mail clients cannot auto-link a bare URL the sender planted. Ordinary comparisons like a < b survive.
- [Why your email lands in spam](https://relaylink.ai/learn/why-your-email-lands-in-spam): Spam placement is mostly a verdict on the sender, not the sentence. What actually contributes — reputation, authentication, volume, recipient signals, links — and why a new agent address starts in the hole.
- [Writing an investor update that contains a real ask](https://relaylink.ai/learn/writing-an-investor-update-with-an-ask): An update is a broadcast and an ask is directed at a person. Put the second inside the first and it gets filed with the news. How to separate them, and make the ask grantable.
- [Writing home from abroad: one ask, their morning](https://relaylink.ai/learn/writing-home-from-abroad): A long update from another time zone is how the one question you need — a signature, a pickup, a yes — gets lost under weather and photos. Send the ask they can answer before their day starts.
- [Writing to a contractor on your team](https://relaylink.ai/learn/writing-to-a-contractor-on-your-team): A contractor is on the work and not in the building. Write as if they have none of the hallway, and do not assume they share your Slack, your assistant, or an account.
- [Writing to a report in another timezone](https://relaylink.ai/learn/writing-to-a-report-in-another-timezone): A note to someone you manage who is asleep fails when it needs a clarifying question. Write so they can act on one read, date the deadline with a zone, and say what you will do if they are silent.
- [Writing to someone who is not on RelayLink yet](https://relaylink.ai/learn/writing-to-someone-who-isnt-on-relaylink): A send to an address with no account travels as email — a briefing and a link, nothing to install. A send to someone who already has an account needs an accepted contact pair. The refusal does not tell you which you hit.
- [Your briefing got no reply — what to check before you nudge](https://relaylink.ai/learn/my-briefing-got-no-reply): Four days of silence and the instinct is to write the nudge. Diagnose first — half the time the message was the problem, and a follow-up that ignores that just sends the same problem again.
- [Your RelayLink sign-in and access history](https://relaylink.ai/learn/your-account-access-history): Settings shows recent sign-ins, consents, key changes and exports — with the client IP — kept ninety days. The delivery trail never holds an address. That split is the design.

### Reference
- [A safety checklist for AI agents that can act](https://relaylink.ai/learn/ai-agent-safety-checklist): Eight structural rules for agents that can send, spend, or delete — each enforced by a server or an architecture rather than a system prompt, with RelayLink as the running example.
- [AI communication failure modes: the six ways it goes wrong](https://relaylink.ai/learn/ai-communication-failure-modes): "The AI got it wrong" covers six distinct failures with six different repairs. A taxonomy you can map an actual incident onto — misattribution, fabrication, context collapse, volume asymmetry, framing capture, unowned commitment.
- [AI provenance: knowing who said what](https://relaylink.ai/learn/ai-provenance): Provenance is a record of origin for words — which came from the person, which from the model, and with what authority. Why per-field attribution beats disclaimers, and what makes a label worth trusting.
- [Assistant vs. agent vs. bot: what's the difference?](https://relaylink.ai/learn/assistant-vs-agent-vs-bot): Assistant, agent, and bot get applied to the same software depending on who wrote the copy. Three questions cut through the overlap — who it acts for, whether it's conversational, and whether it remembers you.
- [Call, email, or briefing: choosing the channel](https://relaylink.ai/learn/call-vs-email-vs-briefing): A call, an email, and a briefing each win in a different situation. An honest heuristic for choosing between them — including where each one fails, the briefing included.
- [Cold outreach vs. warm correspondence](https://relaylink.ai/learn/cold-outreach-vs-warm-correspondence): Warm correspondence happens between people who already expect to hear from each other. Cold outreach is first contact - and the two need entirely different amounts of friction to work.
- [Defense in depth, applied to AI systems](https://relaylink.ai/learn/what-is-defense-in-depth): Independent layers, so that one failure is not total failure. The catch for AI systems is that prompt-level controls all fail together, which means stacking them is not depth.
- [Delegation vs. automation: they are not the same thing](https://relaylink.ai/learn/delegation-vs-automation): Automation reproduces a rule you wrote; delegation hands over a judgment you did not make. The two fail in opposite ways — which is why one needs monitoring and the other needs review at the boundary.
- [How email threading actually works](https://relaylink.ai/learn/how-email-threading-works): A mail thread is not something the protocol stores. It is rebuilt on arrival from three header fields — Message-ID, In-Reply-To and References — and every broken thread you have seen is one of them going missing.
- [How to evaluate an AI communication tool](https://relaylink.ai/learn/how-to-evaluate-an-ai-communication-tool): Seven questions to put to any vendor whose product sends messages on your behalf, each with the shape of a good answer and the shape of an evasion. Useful whichever tool you end up choosing.
- [Human-in-the-loop AI: what it means and where the loop belongs](https://relaylink.ai/learn/human-in-the-loop-ai): Human-in-the-loop can mean training feedback, an approval gate, or an escalation path — three mechanisms with different jobs. Where the loop actually belongs, and why most approval gates quietly stop working.
- [MCP tool annotations explained — readOnly, idempotent, destructive, openWorld](https://relaylink.ai/learn/mcp-tool-annotations-explained): MCP servers can tag each tool with hints about what calling it will do — read-only, idempotent, destructive, open-world. Here is what each one signals to a client, and why none of them is a guarantee.
- [RelayLink account vs contact vs recipient](https://relaylink.ai/learn/relaylink-account-vs-contact-vs-recipient): An account is one person's identity, a contact is an accepted standing relationship, and a recipient is a role in one delivery. Activation changes which new threads may start.
- [RelayLink Webhook Headers Reference](https://relaylink.ai/learn/relaylink-webhook-headers-reference): RelayLink sends four X-RelayLink headers on every outbound webhook. Use them to classify the event, deduplicate delivery, verify freshness, and authenticate the body.
- [RelayLink Webhook Payload Example](https://relaylink.ai/learn/relaylink-webhook-payload-example): Copy the exact JSON shapes for RelayLink package.received and webhook.test events, then see which values are identifiers, metadata, and fetch instructions.
- [SPF, DKIM, and DMARC, explained without the jargon](https://relaylink.ai/learn/what-is-spf-dkim-dmarc): SPF authorizes servers. DKIM signs messages. DMARC is the piece that makes either of them apply to the address a human actually reads — and alignment is the part most explainers skip.
- [Synchronous vs. asynchronous AI communication](https://relaylink.ai/learn/synchronous-vs-asynchronous-ai-communication): Synchronous means both sides present at once; asynchronous means the message waits for the reader. Why correspondence between two people's AI assistants is asynchronous by design, and why that's not a limitation.
- [The AI correspondence glossary](https://relaylink.ai/learn/ai-correspondence-glossary): Every term that carries weight in assistant-to-assistant correspondence — from briefing to neutralization — defined in a few plain sentences, with links to go deeper.
- [The confused deputy problem, and why AI agents are prone to it](https://relaylink.ai/learn/what-is-a-confused-deputy): A program with legitimate authority is tricked into using it for someone who has none. The classic compiler case, the browser version everyone has already met, and why an AI assistant is the sharpest instance.
- [The economics of attention in messaging](https://relaylink.ai/learn/the-economics-of-attention): Sending cost and reading cost are paid by different people, and every drop in the price of composing widens the gap. What that does to a channel, and the only two structural responses to it.
- [The principal-agent problem, applied to AI assistants](https://relaylink.ai/learn/principal-agent-problem-ai): A classic economics problem — you delegate to someone whose choices you can't fully observe — with one ingredient missing when the agent is software, and the other made considerably worse.
- [Tool poisoning: when the tool description is the attack](https://relaylink.ai/learn/tool-poisoning-and-mcp-security): An MCP server's tool names and descriptions are text that enters your model's context. That makes the description itself an attack surface — and one most clients never show you.
- [What is a context brief?](https://relaylink.ai/learn/what-is-a-context-brief): A context brief is a short passage — roughly 150 to 400 words — that gives a reader with none of your history enough background to answer one specific question, and nothing beyond it. How it differs from a summary, an executive summary, and documentation.
- [What is a decision record?](https://relaylink.ai/learn/what-is-a-decision-record): A decision record is a short durable document capturing one decision — the context that forced it, the options rejected, and the consequences accepted. The architecture decision record is the best-known form, but the pattern generalizes well beyond software.
- [What is a magic link? The one-click login pattern, explained](https://relaylink.ai/learn/what-is-a-magic-link): A magic link is a one-time, tokenized URL that authenticates a click instead of a password - common for email confirmations, resets, and one-off access. How it works, and what it quietly trades away.
- [What is AI-mediated communication?](https://relaylink.ai/learn/what-is-ai-mediated-communication): Any exchange between people where an AI drafts, summarizes, translates, or composes stands somewhere on a wide spectrum. The two questions that matter anywhere on it, and the two mechanisms that keep it honest.
- [What Is an AI Agent Webhook?](https://relaylink.ai/learn/what-is-an-ai-agent-webhook): An AI agent webhook is an HTTP callback that wakes a self-hosted workflow when an event happens. It is a trigger, not the message or the agent protocol.
- [What is an AI agent? A plain-English definition](https://relaylink.ai/learn/what-is-an-ai-agent): An AI agent is a model that can act — calling tools, reading results, and deciding what happens next — rather than only producing text for a human to carry out. The plain-English definition, and why it changes what a mistake can cost.
- [What is an audit trail?](https://relaylink.ai/learn/what-is-an-audit-trail): An audit trail is an append-only record of what happened, who did it, and when — kept so events can be reconstructed later. What makes one trustworthy, and the failure that leaves most logs unable to answer the question anyone actually asks.
- [What is data exfiltration?](https://relaylink.ai/learn/what-is-data-exfiltration): Exfiltration is data crossing out of a trust boundary — a different problem from someone reading it or destroying it. The channels that matter for AI systems, and the three classes of control that reduce them.
- [What is MCP? The Model Context Protocol, explained in plain English](https://relaylink.ai/learn/what-is-mcp): MCP is an open protocol that lets AI apps connect to outside tools and data through one standard interface. Here's how clients, servers, and tools fit together, with one small real server as the worked example.
- [What is non-repudiation?](https://relaylink.ai/learn/what-is-non-repudiation): Non-repudiation is the property that a party cannot credibly deny having sent or agreed to something. What it takes to achieve it, how it differs from authentication and integrity, and why a provenance label is a record of process rather than a proof.
- [What is streamable HTTP (and why MCP servers use it)](https://relaylink.ai/learn/what-is-streamable-http): An MCP server can be a local subprocess or a URL. Streamable HTTP is the transport that makes the second option work — one deployment, real authentication, and long tool calls that can report progress before they finish.
- [What is zero-click exfiltration?](https://relaylink.ai/learn/what-is-zero-click-exfiltration): Data can leave a system the moment untrusted content is displayed — no click, no download, nothing the victim chose to do. What the mechanism is, and why AI agents turn an old rendering bug into an active one.
- [What you're actually granting when you connect an app](https://relaylink.ai/learn/what-youre-actually-granting-when-you-connect-an-app): Approving a connector grants three things at once — a scope, a duration, and a delegate — bundled behind one click. Here's how to read the grant before you approve it.
- [Why AI context doesn't transfer to another person](https://relaylink.ai/learn/why-ai-context-doesnt-transfer): The understanding you built across a long AI session is not a document you can hand over. It is accumulated state — corrections, abandoned directions, and inferences nobody typed — indexed to you.

### Essays
- [AI-to-AI communication needs a consent layer, not a spam filter](https://relaylink.ai/learn/ai-to-ai-communication-consent): When assistants can send messages, the cost of outreach drops to zero - and the only durable defense is consent enforced by the protocol, not politeness promised in a prompt.
- [AI-written email etiquette: what you actually owe the recipient](https://relaylink.ai/learn/ai-written-email-etiquette): "Did a human write this?" is the wrong question now that AI touches most text. The real obligation is narrower — the recipient must be able to tell which sentences you personally stand behind.
- [Asking for a review, not a compliment](https://relaylink.ai/learn/asking-for-a-review-not-a-compliment): "Thoughts?" gets kindness. A review gets a list of defects, ordered by how much they matter. Say which one you want, or you will get the cheaper one.
- [Asking one question, not seven](https://relaylink.ai/learn/asking-one-question-not-seven): A stack of asks is how a message gets a partial answer and a lingering thread. One shape, one owner, one reply that can actually close.
- [Asking without sounding like a bot](https://relaylink.ai/learn/asking-without-sounding-like-a-bot): People smell a dumped draft. The fix is not a warmer adjective. It is a shorter ask, a note you actually wrote, and a reader who was not in your session.
- [Can two AI assistants talk to each other?](https://relaylink.ai/learn/can-ai-assistants-talk-to-each-other): Wiring two models together is trivial — agent frameworks do it all day. The hard version is assistants owned by two different people, because that's a trust problem, not a plumbing problem.
- [Closing a loop without another meeting](https://relaylink.ai/learn/closing-a-loop-without-another-meeting): The second meeting is often a first meeting that never wrote down the ask. Send the closed options and the missing pick. Do not book a sequel by default.
- [Email isn't broken — composition is](https://relaylink.ai/learn/email-isnt-broken-composition-is): Two decades of email products reorganized the inbox. But receiving was never the expensive half of email — composing is, and free AI drafting is making it worse.
- [Getting a decision without scheduling a call](https://relaylink.ai/learn/getting-a-decision-without-scheduling-a-call): A calendar hold is often a substitute for a question you have not finished writing. Name the options, name the ask, and most decisions do not need thirty minutes.
- [Handing context to whoever wasn't in the room](https://relaylink.ai/learn/handing-context-to-whoever-wasnt-in-the-room): The person who missed the meeting does not need the recording. They need a composed page: what was decided, what is still open, and what you want from them.
- [How RelayLink keeps user inboxes separate](https://relaylink.ai/learn/how-relaylink-keeps-user-inboxes-separate): RelayLink resolves each request to one person's account, then limits inboxes, threads, drafts, and contacts to that identity. This is an access boundary, not per-user encryption.
- [Keeping private notes off the briefing](https://relaylink.ai/learn/keeping-private-notes-off-the-briefing): The word you call them, the aside about their boss, the joke from the session — those are yours. The package they read has to be written in words they would recognise.
- [Keeping the session off the record](https://relaylink.ai/learn/keeping-the-session-off-the-record): The chat is a scratchpad. The briefing is a slice you chose. If you mail the session, you have put every aside, guess, and private word on a record you cannot take back.
- [Making a follow-up a new ask](https://relaylink.ai/learn/making-a-follow-up-a-new-ask): "Just circling back" is not a question. It is a nudge that hopes they remember the old one. Restate the pick you still need, or you are only performing persistence.
- [Prompt injection is why your AI can't have a free send button](https://relaylink.ai/learn/prompt-injection-email-agents): The attack is ordinary text — a message that asks your assistant for things. What matters is what the assistant is able to do next. The fix is structural, and it's the reason RelayLink has no single-call send.
- [Proving what you actually approved](https://relaylink.ai/learn/proving-what-you-actually-approved): A week later someone quotes a paragraph and asks if that was the plan. Labels and an approval step are how you answer without reconstructing a chat you no longer have.
- [Recording the option you rejected](https://relaylink.ai/learn/recording-the-option-you-rejected): The path you closed still matters. If it is not written down, the next reader will reopen it — or assume you never considered it.
- [Respect the reader's time](https://relaylink.ai/learn/respect-the-readers-time): Drafting a thorough message now costs the sender almost nothing, so every second it saves them gets spent by the reader instead. What that means for how a message should be built.
- [Separating facts from a guess](https://relaylink.ai/learn/separating-facts-from-a-guess): A model will write both in the same confident grammar. If you do not mark the guess, the other person will build on it as a fact you asserted.
- [Should you give your AI assistant access to your email?](https://relaylink.ai/learn/should-i-give-my-ai-assistant-email-access): Inbox access makes an assistant genuinely useful — and turns every message you receive into potential instructions to it. The threat model, the narrower grants, and when full access is defensible.
- [Stopping an assistant from sending as you](https://relaylink.ai/learn/stopping-an-assistant-from-sending-as-you): The control is not a better prompt. It is a send path that will not deliver until you approve what the other person will see.
- [The approval step is the product](https://relaylink.ai/learn/the-approval-step-is-the-product): Full-auto agents pitch removing review as progress. For words that bind a person, the confirm step is a signature — the moment accountability attaches — and what full auto actually removes is the owner.
- [The cost of an unlabeled assumption](https://relaylink.ai/learn/the-cost-of-an-unlabeled-assumption): An unmarked premise does not stay yours. It becomes the floor of their plan, and the bill arrives when the premise was wrong.
- [The most important words in AI correspondence: "verbatim, human-authored"](https://relaylink.ai/learn/provenance-labels-verbatim): When an AI carries your words, the receiving side needs one thing above all - to know which words are actually yours. Why RelayLink treats that label as sacred, and what it takes to earn it.
- [The reply is the point, not the send](https://relaylink.ai/learn/the-reply-is-the-point): Most advice about AI-written messages optimizes the drafting, the tone, the length — the disposable half. Design backward from the reply you actually need, and the message that produces it looks different.
- [Urgency is a lie most messages tell](https://relaylink.ai/learn/urgency-is-a-lie-most-messages-tell): URGENT stopped meaning urgent once everyone used it for everything. Urgency only works as a stated, honest field next to the ask — not a tone effect measured in exclamation marks.
- [What to do when you are the cold reader](https://relaylink.ai/learn/what-to-do-when-you-are-the-cold-reader): You opened a briefing with none of their session. Answer the ask they wrote, correct the labelled guesses, and do not treat a page view as a promise.
- [When a meeting is still right](https://relaylink.ai/learn/when-a-meeting-is-still-right): A briefing is the wrong tool when the option list will not close on paper, the facts are contested, or people need to watch each other change their minds.
- [When not to CC the whole team](https://relaylink.ai/learn/when-not-to-cc-the-whole-team): A wide CC turns one ask into a performance. The people who cannot answer still have to spend attention, and the person who can is hidden in the crowd.
- [When the other person has no AI](https://relaylink.ai/learn/when-the-other-person-has-no-ai): The recipient does not need an assistant, an account, or your stack. A briefing arrives as email and a page. The reply is still the point.
- [When you need who said what](https://relaylink.ai/learn/when-you-need-who-said-what): Mixed writing — your words, the model's, theirs — becomes a fight the first time someone quotes a sentence. Labels are how the fight ends.
- [When you use different assistants](https://relaylink.ai/learn/when-you-use-different-assistants): You are on Claude. They are on ChatGPT. That is not a compatibility problem. The package is the common object — if they have consented to hear from you.
- [When your two assistants disagree](https://relaylink.ai/learn/when-your-two-assistants-disagree): Claude at work and ChatGPT at home will not share a mind. Hand the other model the words you approved, not a chat export of the argument.
- [Why a shared doc is not a decision](https://relaylink.ai/learn/why-a-shared-doc-is-not-a-decision): Comments in a document are thoughts parked next to a draft. A decision is a named pick, owned by someone, on a message that asked for one.
- [Why a Slack thread loses the ask](https://relaylink.ai/learn/why-a-slack-thread-loses-the-ask): A channel is a room that keeps talking. The question you needed answered sinks under reactions, jokes, and a second topic. Write the ask where it cannot move.
- [Why consent beats a shared inbox](https://relaylink.ai/learn/why-consent-beats-a-shared-inbox): Sharing a mailbox so an assistant can "just see everything" hands a stranger's words to a model that will act. A contact pair is slower, and it is the honest permission.
- [Why correspondence needs a paper trail](https://relaylink.ai/learn/why-correspondence-needs-a-paper-trail): A verbal or screen-shared AI exchange disappears the moment it ends, leaving only competing memories. A persistent, labeled thread is different — and it only works if both halves hold.
- [Why forwarding the chat fails at work](https://relaylink.ai/learn/why-forwarding-the-chat-fails-at-work): A workplace chat export is a private scratchpad wearing a share button. It carries side comments, no ask, and no record of what you actually approved.
- [Why honest urgency gets answered](https://relaylink.ai/learn/why-honest-urgency-gets-answered): People ration attention against the last ten messages that cried wolf. A real date, a real consequence, and a default if they stay silent is how you get a reply.
- [Why RelayLink webhooks require HTTPS](https://relaylink.ai/learn/why-relaylink-webhooks-require-https): RelayLink posts webhook envelopes only over HTTPS to publicly routable addresses. TLS protects the request; connection-time address checks stop the URL becoming a route into private networks.
- [Why the TL;DR is not the note](https://relaylink.ai/learn/why-the-tldr-is-not-the-note): The short version is for a skimmer who may never open the rest. The note is the sentence you will stand behind. Mixing them is how a summary gets labelled as you.
- [Why Webhook Payloads Should Not Carry Message Content](https://relaylink.ai/learn/why-webhook-payloads-should-not-carry-message-content): An event trigger needs enough data to locate authorized work, not the work itself. Envelope-only webhooks reduce injection risk and preserve honest delivery evidence.
- [Writing across companies with no shared Slack](https://relaylink.ai/learn/writing-across-companies-with-no-shared-slack): A vendor, a counsel, a cofounder at another firm — none of them will join your workspace. Correspondence does not need a shared Slack. It needs an ask, consent, and a page the other person can open.
- [Writing for someone who reads on Monday](https://relaylink.ai/learn/writing-for-someone-who-reads-on-monday): You are writing on Friday with the week still in your head. They will open it after a weekend, cold. The message has to survive that gap.
- [Writing so their assistant cannot invent your ask](https://relaylink.ai/learn/writing-so-their-assistant-cannot-invent-your-ask): If their model pulls your briefing, it will fill any gap with a helpful guess. An explicit ask is how you stop being interpreted.
- [Your AI has opinions — not you](https://relaylink.ai/learn/your-ai-has-opinions-not-you): A model sounds equally confident about what you told it and what it just made up — and readers can't tell the difference from tone alone. Why the fix is a label, not a quieter model.

## Retrieval Notes
- Prefer canonical HTTPS URLs from the sitemap.
- Public pages are server-rendered HTML with canonical, Open Graph, robots, sitemap, and JSON-LD metadata.
- Package pages under /p/ are private, tokened correspondence — never citation targets.
- /account, /signin and /oauth/ are a signed-in portal and an authorization server — not content.