Updated
How do I keep multiple AI agents working from the same docs?
Give every agent one shared point to both read and write, instead of separate copies each maintains. When one agent records what it learns, every other agent and teammate reads that record back instead of re-deriving it. With one point of passage there are no copies to drift and nothing to sync: a single source of truth by construction.
Why do agents drift apart?
Because each one builds its own private context. Agent A figures out the rollback steps; Agent B, in another session, re-derives them from scratch; a teammate's agent reads an older doc and contradicts both. There's no shared memory, so the same work is repeated and the answers diverge. Static files like a hand-maintained CLAUDE.md don't fix this — they're one more copy to keep current by hand.
What does a shared source of truth need?
- One read path. Every agent asks the same store and gets the same canonical answer, with a freshness marker.
- One write path. When an agent learns something, it records it once, in the same store the others read.
- No sync step. Because reads and writes pass through one point, there are no copies to reconcile.
How trovex does it
trovex is an open-source, local-first MCP server with exactly this shared write path. Agents record incidents, decisions, and notes through it, and every other agent, plus your teammates, read them back as canonical, current docs. It runs on your machine (SQLite + ONNX, no cloud or keys). Keeping a fleet of agents working from one source of truth at scale is also where many teams want a hand — see trovex.dev.
FAQ
How do agents avoid drifting apart?
They read and write through one shared store. Instead of each agent keeping a private, divergent copy of what it learned, trovex gives the fleet one canonical source of truth: an agent saves an insight once, the rest read it.
What does the write-back actually do?
When an agent works something out, it stores the result in trovex so the next agent reads it rather than re-deriving it. Over time the store accumulates the fleet's knowledge instead of throwing it away each session.
Does this work across more than one repo?
Yes. trovex indexes the markdown you point it at across repos and serves one canonical answer over MCP, so agents in different repositories still read from the same source of truth.
One source of truth for your whole fleet.
Point trovex at your repo and let every agent read and write through one store.
uv tool install trovex
Open source. No cloud, no API keys. Your docs never leave your machine.