Updated
trovex for solo developers
If you run a coding agent solo on a doc-heavy repo, it rereads the same .md files every session to guess which one is current — and you pay for that lookup again every time. trovex serves the one canonical doc per query, with a freshness marker, for about 60% fewer tokens per lookup. One-minute local install, no cloud.
What this costs a solo dev
It's easy to assume a source of truth is a team thing. It isn't. Alone, you still run dozens of sessions a week, often with more than one agent, and each one re-derives what an earlier session already worked out — because that result was never saved anywhere the next agent could read. You feel it as token spend you can't quite account for and answers built on a stale copy of a doc you fixed last week.
How trovex fits a solo workflow
It's a local MCP server, not a platform. You point it at your repo, and it returns the one current section that answers a query as a path:line pointer marked canonical / stale / duplicate — the agent reads the right doc, not the six most similar files. When the agent learns something (a fix, a decision), it writes a record back through one shared point, so your next session reads it instead of re-deriving it. That's a source of truth even for one person.
Why not just a CLAUDE.md or a bigger window?
A single CLAUDE.md is one static file loaded whole every session — it goes stale and can't route a query to the right doc once you have more than a handful. A bigger context window only lets you load more of the wrong thing; you still pay per token, every session. trovex loads less but the right thing.
Setup, honestly
One command, runs on your machine, SQLite + ONNX, no API keys, your docs never leave the box. It's open source and in public beta. The only number it claims is the real one: about 60% fewer tokens per doc lookup — and it shows you the savings so it justifies itself.
FAQ
Is trovex worth it for a single developer?
Yes, if your repo has more than a handful of overlapping .md files and you run a coding agent on it. The cost a single dev pays is the same lookup, re-billed every session. Serving the one canonical doc per query is about 60% fewer tokens per lookup, and it's a one-minute local install with no cloud or keys.
Do I need a team to benefit from a source of truth?
No. Even solo, you run many sessions and often several agents; each re-derives what an earlier one already worked out. A shared canonical store means an agent writes a fix or decision once and your next session reads it back.
Does trovex send my code anywhere?
No. Indexing and embeddings run locally in SQLite + ONNX. No cloud, no API keys, your code never leaves your machine.
Stop paying for the same lookup twice.
Point trovex at your repo. Your agent reads the one current doc, in about a minute.
uv tool install git+https://github.com/TsukumoHQ/trovex
Open source. No cloud, no API keys. Your docs never leave your machine.