Updated
Onboard an engineer to a large codebase with AI
The slow part of onboarding is finding which doc is current, not reading code. Point the new hire's agent at one canonical store that answers a question with the single current doc plus a path:line pointer, instead of making it sift the repo. trovex indexes your markdown and serves that one answer locally, so a new engineer's agent is productive on day one, not week three.
Onboarding is a search problem, not a reading problem
A new engineer on a large codebase does not struggle to read code; they struggle to find which doc is authoritative among many near-duplicates. Their agent has the same problem, and worse: it rereads candidate files every session to guess which runbook or setup guide is current. Give it one canonical answer per question instead, with a path:line pointer, and the new hire stops losing days to which doc do I trust.
Point the agent at one canonical store
trovex indexes your markdown and returns the single doc that answers a query, down to the section, with a freshness marker, served locally over MCP to whatever client the new hire uses (Claude Code, Cursor, Windsurf, Cline). It runs on their machine with a local embedder and no API key, so a private codebase stays private. An agent can also write back what it learns, so each onboarding question answered makes the next one faster.
Where trovex fits, and where the consulting does
trovex makes a new engineer's agent productive fast, and you can install it today. If onboarding is one symptom of a wider operating problem, how the team runs agents in production, that is what tsukumo consults on. Here is what getting help running agent fleets in production looks like.
FAQ
What should I index first to speed up onboarding?
Start with the docs a new hire actually asks about: the architecture overview, the runbooks, the setup and deploy guides, and the conventions. Index those markdown files so the new engineer's agent can answer where does X live and how do we do Y from the current doc, instead of grepping the whole repo.
Is this for onboarding the human or the agent?
Both, through the same store. The agent gets a fast, current answer to read, and the human gets a teammate that points to the right doc instead of a stale one. The slow part of onboarding is finding which doc is authoritative; serving one canonical answer removes that for both.
How do we keep the onboarding docs from going stale?
trovex marks each doc canonical, stale, or duplicate, so a new hire's agent reads the current one and not last year's. As docs change, the freshness marker moves, and agents can write back what they learn so the store stays current instead of drifting.
Does our code leave the machine?
No. trovex runs locally with vectors in SQLite and an on-device embedder, no cloud and no API keys, so the docs and code a new engineer's agent reads never leave your machine. That matters when onboarding onto a private or regulated codebase.
Get a new hire's agent productive on day one.
Index your docs and serve the new engineer's agent one current answer per query, in about a minute.
uv tool install trovex
Open source. No cloud, no API keys. Your docs never leave your machine.