Better signal
Live, structured, user-maintained context — not stale forms or siloed copies.
For services & developers
Users hold structured, live personal context. You integrate via computation contracts — query in place, return approved results, compete on intelligence not database size.
Why integrate
Offer better services with user-permissioned data — without storing the liability yourself.
Live, structured, user-maintained context — not stale forms or siloed copies.
Query in place under contract. No warehouse to breach, erase, or audit.
Integrate once via manifest. Skip duplicate collection, storage, and DPO overhead.
Purpose, space scope, permission tier, and consent ledger — user approves every relationship.
The shift
Give us all your data
Permanent copy in your warehouse. Stale the moment it lands. Liability on you.
Run this approved computation
User approves purpose and space. You query live at source. Return the result — optionally write insights back. Not the vault.
How it works
The user is the root database. You are a temporary processor — query live, deliver value, optionally write back.
Approve
User picks counterparty, purpose, spaces, and permission tier.
Read
Partner queries scoped context live — manifest + query API, no warehouse copy.
Deliver
Your product returns value: recommendation, eligibility, automation.
Write back
Optional insights compound the user's graph under delegated write.
Audit
Every read and write visible in Exchanges and the consent ledger. Revoke anytime.
Users can test the loop with the built-in Autograf demo: Try the demo exchange.
Integration paths
User creates a contract with counterparty, purpose, space, and permission tier. You receive manifest and query URLs.
Contracts in dashboard →GET /api/share/[token]?format=manifest returns capability document. POST /api/share/[token]/query runs scoped computation.
Shares in dashboard →Connect Cursor, Claude Code, or custom agents via OAuth-scoped MCP grants. Activity lands in the user's graph under contract.
MCP in dashboard →Full bidirectional loop: query scoped context, deliver value, optionally write insights back. Users see both directions in Exchanges.
Exchanges in dashboard →Developer quickstart: autograf/docs/integrating-services.md in the repo · Concept: concepts/computation-contracts.md
Use cases
Read travel history and calendar under family space. Suggest options — never sell the profile.
Query financial space for a computation. Return yes/no — not full transaction history.
Read work space for project context and hours. Prepare meetings without uploading everything.
Cursor, Claude Code, and custom agents connect via scoped MCP grants.
Read dietary space for meal planning; write lab summary back as a Note under delegated write.
Permission tiers
Map technical permissions to tiers your users can approve with confidence.
Local only
Data stays in the user's space. Read and query — nothing exported.
Private compute
External AI can answer questions. Raw data is not stored outside the space.
Approved export
Specific outputs can leave. Query plus controlled export.
Temporary access
Expires automatically. Good for trials and one-off checks.
Delegated write
Can write insights back into the user's graph with provenance.
Sensitive data
Health, finance, or legal spaces — extra confirmation required.
User-side first
Single-player value — private AI memory — creates the structured graph companies later query.