Updated
What's the single source of truth for a multi-agent repo?
One store that every agent both reads from and writes to, so the repo's documents, not each agent's private context, are authoritative. With one point of passage there are no copies to drift and nothing to sync, and a query returns the one canonical doc with a freshness marker, not whatever an agent cached. It stops fleet cost from compounding.
Why "the repo" needs to be the unit, not each agent
In a multi-agent setup, the failure mode is divergence: each agent builds its own private context, so the same question gets re-derived, and two agents (or two teammates) end up acting on different copies of the same doc. If the source of truth lives inside an agent or an editor, it doesn't travel. Anchoring it to the repo's docs, one store all agents share, is what keeps everyone current.
What the pattern requires
- One read path: every agent asks the same store and gets the same canonical answer, with a freshness marker.
- One write path: an agent records a decision or incident once, in the store the others read.
- No sync: because reads and writes share one point of passage, there are no copies to reconcile.
- Cost that doesn't scale with the fleet: the answer is computed once and reused, not re-derived per agent.
How trovex implements it
trovex is a local-first MCP server with exactly this shared write path: agents read the one canonical doc per query and write records back through the same store, so the whole fleet and your teammates see one current source of truth. For the step-by-step version, see how to keep multiple agents on the same docs.
FAQ
One index per repo, or one shared one?
Either, depending on how your teams work. trovex can index multiple repos' markdown and serve a single canonical answer over MCP, so agents working across services read from one source of truth instead of each repo's local, possibly stale, copy.
How does it know which repo's doc is current?
The same freshness mechanism applies across repos: each doc is marked canonical, stale, or duplicate, so the agent gets the current one regardless of which repo it lives in.
Does pointing it at many repos slow it down?
No meaningfully. It runs locally with vectors in SQLite and an on-device embedder; lookups stay fast because it returns one canonical doc per query, not a scan of every repo.
One source of truth for the whole fleet.
trovex is open source and in public beta. Install it and put every agent on one store.
uv tool install trovex
Open source (AGPL-3.0 core, MIT CLIs). Local-first — your docs never leave your machine.