The core product: open-source memory for AI agents. Primitives are Workspace > Peer > Session > Message, with three background agents — Deriver (forms memory), Dialectic (recall), Dreamer (consolidates).
Start here
- ADR — the architectural decisions, especially the Peer Paradigm.
- Modes of operation — how Honcho is used in practice.
- End-to-end data flow — a worked message→reasoning example.
- Specs — what’s being designed and shipped.
Deeper
- Long memory · Deriver performance · Deriver internals
- CloudEvents — the event/telemetry pipeline and how to query it.
- SDK release workflow · Docs proxy
- Harness plugin testing — sandbox vs. throwaway VM, and which tests run where. Procedure: VM pass runbook.
- Integrations: Vercel AI SDK (revisions).