Push it once.Every teammate’s agent knows it.
ilolink is a registry for what agents produce — skills, specs, plans, runbooks, handoffs. Push once from any repo; every teammate’s assistant reads the same versioned, reviewed set. And anything worth showing the world becomes a link anyone can open, no sign-in to read.
Free for one person. $9 once — not monthly — for a team of five.
One connector, both halves
You say it in chat. It lands where the team can use it.
Add ilolink as an MCP connector in Claude, Claude Code, ChatGPT or any MCP-compatible assistant. One approval, no API key. You pick the teamspace it may write into — fixed for the life of the connection.
“Save this spec for the team.”
spec/checkout-refunds v3 — pushed to Acme teamspace, awaiting admin approval
“Check what we already have on commit conventions.”
skill/commit-style v5 by Dana, 12 Jun — read before starting
“Publish this as an ilolink page.”
ilolink.com/q3-pricing — link back, analytics behind it
“Save the handoff as a session transfer.”
session/billing-migration v1 — the next person picks up what was decided
The registry
One name means one thing, for every agent on the team.
01
Connect an assistant once
Add ilolink as an MCP connector in Claude, ChatGPT, or your coding agent. You pick which teamspace it may write into — fixed for the life of the connection.
02
Push from any repo
The file's path already says what it is — .claude/skills is a skill, docs/decisions is a decision. Every write is a new version with an author and a note.
03
Teammates pull the same set
Anyone fetches it by name, from any repo, on any machine. What comes back names the author, the version and the date.
04
Review before it's policy
A member's push arrives as a proposal. An admin approves or rejects it — before anyone's agent starts obeying it.
Reusable instructions for a recurring task. Read at the start of work to see whether the team already has guidance.
A subagent definition: its role, the tools it may use, and how it should behave.
What is being built and why, including the behaviour an agent should implement. The agreed shape of the work.
How something is built: architecture, data model, and the trade-offs that were weighed.
An ordered implementation plan: phases, dependencies, and what 'done' means for each step.
A multi-step orchestration an agent runs: the stages, what fans out, and what verifies.
A handoff from one working session to the next: what was done, what was learned, what is still open. Read this to resume someone else's work.
A decision that has been made and should not be silently reopened, with the reasoning that produced it. The answer to 'why is it like this'.
An operational procedure to follow exactly: deploys, migrations, incident response, rollbacks.
Cases an agent's output is checked against: the input, the expected behaviour, and what counts as a failure.
A registry entry isn’t a document someone might read — it’s an instruction another person’s assistant will act on. So a member’s push arrives as a proposal, and an admin approves it before any agent obeys it. Teams that don’t want the step turn it off.
The other half
Put a document online, and find out what happened to it.
Paste Markdown or HTML, or drop in a PDF or .docx. One link anyone opens without an account — and a private page showing who read it, how far they got, and what they wrote back.
Write first, make the account at the end — your draft stays where it is while you sign in. Choose who can open it: public, unlisted, password-protected, or expiring on a date.
Revise in place and the link stays the same, previous version kept. Nobody ends up reading the wrong file.
Publishing needs an account. Reading never does.
Views, unique readers, and the share who reached a quarter, half, and the end. Plus referrers, countries, devices, and time on the page day by day.
“This is the paragraph everyone reread — can we say the price here, not on page 4?”
Anchored comment · 2 replies · 👍 4 · 🤔 2
Heatmaps show where readers clicked and where they stopped. Comments sit against the passage being discussed — anonymous by default, named if you ask, off if you’d rather.
One honest limit: a PDF opens in the browser’s own viewer, so read-depth and heatmaps can’t see into it. Views, sources, reactions and comments still work on a PDF.
Who reaches for it
Team lead standardising agents
Agent definitions and runbooks in one reviewed place — how an agent behaves is something the team decided, not one person.
Engineer with a house style
Push commit conventions once. Every teammate's assistant reads the same rules in every repo.
Anyone ending a long session
Save the handoff as a session transfer. The next person — or model — picks up what's decided and what's open.
Consultant sending a proposal
The PDF as a link, not a 12 MB attachment. See whether the client opened it at all.
PM circulating a spec
Unlisted link, comments anchored to the disputed paragraph, revised in place — the link never changes.
Founder sending an investor update
One expiring link. Views, read depth, who came back — no tracking pixel, no cookie banner.
Pricing
Pay once, or don’t pay at all.
Working alone is free. A team plan is a one-time payment — you pay for the team size once and keep it. Nothing recurs, nothing expires, no card on file.
Personal
Free
For one person, publishing a few documents.
3 published documents
Just you — invites need a team plan
Full analytics: views, read-through, heatmaps
Comments and reactions from readers
Connect your AI assistant over MCP
Team of 5
$9
Pay once. Five people, forever.
100 published documents
5 teammates, including you
Shared registry: skills, specs, plans, handoffs
Everything in Personal
Team of 10
$19
Pay once. Ten people, forever.
500 published documents
10 teammates, including you
Shared registry: skills, specs, plans, handoffs
Everything in Personal
Personal is one person: you. Inviting anybody — a teammate, a co-founder, one reviewer who writes to the registry — takes a team plan. That’s the only thing the payment unlocks, and you make it once.