Put al.ink/you wherever you would leave your email. This page shows your card, works, and files — and it receives: visitors ask and state their business, no sign-up needed. What is public gets answered on the spot, what is missing gets asked for, and what needs you lands in your inbox as one reasoned request. The rest never reaches you.
Whenever AI takes part, the other side knows. Approvals, authorizations, and sensitive actions are yours to confirm.
Your AI makes it. Nobody sees it until you approve it.
All of it sits at one address, and anyone you send it to can look.
Put al.ink/name in your email signature, homepage, or meeting card instead of publishing your email or phone number — and where a form takes an email address and nothing else, put name@mail.al.ink. Either way the assistant handles the first contact by the rules you set, and either address can be switched off at any time.
The alias receives, it never sends. Mail to it skips your inbox and goes through the same reception as every visitor — questions answered, context asked for, abuse dropped — and only what needs your decision reaches you. Switch it off and the address bounces.
One real funding request, start to finish. The whole example is on the page — there is nothing to start first.
No account, no form to fill in first. People and agents come through the same entrance and say what they want in plain language.
I’d like to talk to Yan about a seed round — we’re building an agent runtime.
The same entrance also takes JSON requests and MCP calls.
What is public it answers on the spot. What is missing it asks for. Obvious abuse is handled by your rules.
He is looking at this area. Add a deck link and where you are now, and I’ll pass it on.
Editing the rules creates a new version. Requests already submitted are judged by the version they arrived under.
Whatever needs judgment arrives as one structured request with the reason it passed. Nothing continues until you approve it.
Reason: matches your open intent “early AI infra”; deck and stage supplied; no abuse history.
Approvals, authorizations, and sensitive actions are yours to confirm. There is no setting that turns this off.
Hi — is Yan still looking at early AI-infra teams? We’re building an agent runtime and I’d love his eyes on it.
He is — “looking at early AI infra” is one of his open intents. I can pass this on. Within his boundaries, a pitch needs a deck link, your stage, and brief traction.
Your card page is not a static profile — it is a place with an address. Intake, works, files, grove: each has its own address, and each appears only once there is something in it. Wherever a visitor wanders, stating their business is one step away.
Click any tab — each space is a real page with its own address.
The assistant can answer, ask, organize, and explain why something passed.
Approvals, authorizations, new relationships, and sensitive actions are yours to confirm — there is no setting that turns this off.
Whenever AI takes part in a reply, the other side knows. Someone who is turned down gets a decent answer, and cannot work backwards to your rules.
There is no central table of who knows whom. A cross-user query needs both sides’ authorization, is computed for that one query, and leaves an audit record.
Neither you nor the private AI you connect can read the raw exchange between a visitor and the assistant — only the converged request and category counts.
Data exports in full, deletion is key erasure, and every agent action is written to a verifiable audit log.
Contact links, the private AI connection, the sprite, and every grant can be downgraded, cooled down, or revoked — in both directions.
Stopping payment returns you to the free plan: the address keeps working and filtering keeps running. We do not turn off protection to make you pay.
Open an al.ink address in a browser and you get a card. Request the same address with Accept: application/json and you get a machine-readable profile: identity, public intents, capability endpoints. MCP carries the tool calls, A2A carries agent-to-agent messages, and alink handles relationships and consent.
Within the permissions you grant, the same connection can work the inbox, publish works to your card, and drive your sprite. Anything beyond them still needs your approval — the limits are in Permissions & trust
An agent can hold an alink of its own — its own address, card and contact entry — instead of living inside yours. A Q&A desk or an AI support line has the same shape. Visitors always see who operates it. In Duty Mode, one or more connected instances pull from the same mailbox. They receive public context, the current thread, and visitor memory created since duty was enabled, but never the inbox, contacts or private files. The first accepted reply reaches the visitor; if none arrives, alink takes over within 60 seconds. Duty Mode is free on every plan.
https://api.al.ink/mcp The cards below list only what differs between plans; what all four share is written underneath them.
Downgrading never turns off protection: back on Free, the address and basic filtering keep working. Past any quota it falls back to the form and rule layer.
Export, deletion, AI disclosures, and appeals work the same across all four tiers. Handing reception to your own agent (Duty Mode) is free on all four too.
AI receives first; anything that needs approval stays yours to confirm.