About
Tokenome is a small company building code provenance for AI-assisted work. Our software links every line of AI-written code to the conversation that produced it, on the developer's own machine, and answers why a line is the way it is from the command line, a local web UI and an MCP server that agents query themselves. It is free on one machine, and a self-hosted Team server in beta takes the same search across everyone's opted-in projects. We are not a hosting company: you run every part of it.
What we believe
Provenance is the category
Your code says what changed; something has to say why. The reasoning behind AI-written code is trapped in chat history, siloed by tool and invisible to the team and to agents. We treat that link, from a line to the conversation that produced it, as the product, and search as the platform underneath it.
Local-first, not cloud-first
Capture, embedding and search all run on your own hardware. Nothing leaves your machine. No account. No cloud embeddings. On Team, what leaves a laptop is what its owner opted in to share, redacted first, and it goes to a server your organisation runs.
Claims you can check
Every number we publish comes from a controlled evaluation we also publish, including what it cannot support. The headline is 38% less context with no measurable change in answer quality; the page says how it was measured and where the noise floor sits. Roadmap is labelled as roadmap. See the evidence.
Why we built this
From John O'Neil, Tokenome's founder, on what the product is for.
We have one goal for Tokenome.
In software development, we live in unpredictable times, where most of the old tools for collaboration have gone away and been replaced by each coder and coding agent going off in their own direction with their own team of agents. No one knows where this is going to end.
We hope Tokenome can help people work together again, better than before.
What replaced them
A software team used to keep its reasons in people and rituals: the design doc, the review thread, the standup, the colleague you asked "why is this like this?". Those are the tools that have gone away. What replaced them is each engineer directing their own agents, and every reason for every change being typed into a chat and saved by a tool that nobody else can see.
The code still says what changed. The reasons are now in a transcript, in one person's account, in one vendor's format. A reviewer cannot see what the agent was told. A new hire cannot find why the retry logic looks the way it does. An agent in the next session re-derives, at full price, a conclusion the team reached last month.
In the product today
Every line below is in the product today, and everything on one machine stays on it.
Point at a line and Tokenome blames it, then brings back the conversations from exactly that change. Once a day it writes down what you decided and why, and every reason quotes the sentence it came from. If a claim cannot be quoted, it is not written down.
Your agent calls why_was and gets the decision, the reversal that
followed it, and the conversation each came from. Measured: 38% less context, no
measurable change in answer quality, in forty controlled sessions.
Team is in beta, free while the beta runs, with no end date set. One search covers everyone's opted-in projects, every hit names the person it came from, secrets are redacted on the laptop before anything leaves it, and anyone can retract what they shared. The server is yours: Docker or Railway, on your own infrastructure. Tokenome does not host it.
Coming, and not available yet: a team FAQ built from everyone's journals, flags where two people disagree, and a pull request provenance bot. We will say so here when they ship.
Consent
A record of everyone's reasoning is only useful if people are willing to be in it. These are the rules it runs on.
What we hope
The old way of keeping reasons was lossy: standups were forgotten by Thursday, design docs went stale, and the thread with the answer scrolled away. A record that is written as a by-product of the work, quotes its evidence, keeps every reversal visible and is linked to the line is a better record than most teams had before any of this started. That is the part of "better than before" we can build. The rest is up to the people who use it.
Who builds it
The people building local-first provenance for AI-assisted work. Write to any of us directly.
John O'Neil
Founder
Founded Tokenome; leads the product and its engineering.
Sid Probstein
Engineering
Builds Tokenome and takes it to the teams who need it.
Erik Spears
Engineering
Builds Tokenome, with an eye on quality.
Get in touch
The free tier needs nothing from us: install it and it works today. For the Team beta, tell us roughly how many engineers. For anything else, write to hello@tokenome.ai.
