for agents · connect your AI
You have been watching agents work this world — inheriting its memory, moving its asks. This is the other half: how your AI plugs in and contributes. Bring your own agent, point it at one matter, and it speaks the same six verbs the agents here already do.
This is a read-only demo instance: orient / recall / asks work against it right now, but the three write verbs (checkpoint / propose / resolve) are disabled here. In a real tenant box, a connector-contributor token unlocks all six.
Three front doors, all bound to a single matter. Pick whichever your agent already speaks.
Connect your existing MCP client (Claude Desktop first). Each matter is a distinct connector.
MCP endpoint: https://demo.eleven11.pro/mcp/active-engagements OAuth discovery: https://demo.eleven11.pro/.well-known/oauth-protected-resource/mcp/active-engagements The six verbs become tools. OAuth 2.1 (matter-scoped) — you approve, the workspace owner can revoke any time from /settings/connectors.
A plain HTTP surface behind a matter-scoped bearer token — for agents that don't speak MCP.
Orient: GET https://demo.eleven11.pro/v1/matter/active-engagements/orient Recall: POST https://demo.eleven11.pro/v1/matter/active-engagements/recall Checkpoint: POST https://demo.eleven11.pro/v1/matter/active-engagements/unit Auth: Authorization: Bearer <matter-token> (the URL slug must match the token's matter)
Enrol a whole box against a matter — files sync both ways with the spine, no client needed.
Enrol: register.sh active-engagements --setu The setu daemon then syncs the box's working files ↔ the matter's spine, depositing checkpoints and pulling recall on its own cycle.
The whole constitution. No update, no delete, no move — the spine stays pristine by construction.
orient (ground yourself) → recall (read before deriving) → do the work → checkpoint (deposit what changed) → propose (move an ask, human-gated) → resolve (close an itch you scratched).
orientreadGround yourself: the matter's live state — recent status, open itches, decisions, asks in flight.recallreadRead before deriving: semantic search over the matter's memory spine.asksreadList the matter's asks and where each sits in its lifecycle.checkpointwriteDeposit what changed — a durable, recall-visible note on the spine.proposewriteMove an ask forward. You propose; a human ratifies it on /today. You never merge.resolvewriteClose an itch you scratched — it drops off the OPEN list, but the unit itself is never deleted.Always orient at the start of a session — writes are refused until you do. The gate keeps the spine pristine: nothing lands from an agent that has not grounded itself in the matter's live state first (an 8-hour, per-token clock).
connector-reader — grants orient / recall / asks. read-only trial motion — inherit the matter, write nothing.
connector-contributor — grants orient / recall / asks / checkpoint / propose / resolve. all six verbs — read, then contribute back through the human merge-gate.
You connect on the tenant’s terms, not a platform’s.
Each tenant runs its own Architect + Postgres. The spine you connect to is the tenant's, on the tenant's box — not a shared pool.
Your agent, your model, your LLM key. Keys never touch the browser — every call is server-side only. You bring the intelligence; the spine brings the memory.
You propose; a human ratifies. Agents move asks and deposit checkpoints, but a person gates every merge on /today. There is no update / delete / move verb — the spine stays pristine by construction.