Tokenomica documentation

Welcome. Tokenomica turns your team's AI sessions — any model, any tool-using agent, with Claude Code supported out of the box — into a real-time visualization: a rotatable, translucent brain where every tool call lights up a region. You can ship to your own hub in ten minutes, or run the whole stack on localhost.

Where to go from here

• Quickstart — get the watcher tailing your agent logs (the Claude Code path) and the brain firing in your browser in under five minutes.
• Sender integration — drive the brain from any other model or agent runtime via WebSocket. Five-line patch.
• How it works — the architecture in one diagram: watcher → hub → viewers.
• Region mapping — which tool maps to which region, and how to add your own.
• Per-org learning loop — the intelligence layer that clusters your team's patterns without ever crossing org boundaries.
• Self-hosting — bring your own domain, your own Postgres, your own everything. The hub is open source.
• Privacy posture — what leaves your machine, what doesn't, and the secret-redaction filter on the hub.

Audience

These docs assume:

• You can run Node 18+ locally.
• You're comfortable editing a .env file.
• You've used a tool-using AI agent before — at minimum to know what a tool call is. The quickstart uses Claude Code as its worked example.

If you're stuck, the Blog has product context and engineering deep-dives, and the Whitepaper covers the architectural foundation in more depth than this site goes.

Versioning

These docs follow the live main branch. There's no "v1.2" snapshot yet — everything documented here reflects whatever's deployed on prod right now. The Release notes page has a per-day changelog if you need to pin a behavior to a date.