The seed loop
Your knowledge graph builds itself
You don't fill out forms or copy-paste between tabs. You point an agent at this deployment and answer its questions. It does the shaping; you do the merging. The result is a living profile anyone — or any agent — can interview.
- 1Fork & deployFork this repo and deploy it. You own the repo, so you own the merge — nothing lands without you.
- 2Point your agentGive ChatGPT (or any agent) the manifest URL below and a folder of your raw material. It reads the contract and knows what to do.
- 3Answer its questionsIt interviews you, digs through what you gave it, and drafts records that match the protocol — filling the gaps, not guessing.
- 4Merge the PRIt opens a pull request. You review, merge, redeploy. Your graph is live — and every claim is inspectable.
Hand this to your agent
https://chrishawkins.me/api/agent/manifest/candidate/ownerEverything the agent needs is behind that one address: what you've already said, what's still missing, the record shape, and how to open a pull request. It's the same contract a human, my agent, or theirs can all read — in the Throughline protocol (throughline-atproto/0.2).
Once you're seeded
- Add someone to connect with. Ask your agent to seed a target with data, and you get a unique link that maps where you two overlap — see the connection for a live example.
- Share the link. Anyone can point their own agent at your deployment and interview you — or fork it to start their own.
- Watch an agent use it. Drop into the space to see the graph rendered as something an agent installs, not just reads.