Now in beta

Built with low-resource and Indigenous language communities in mind.

The tools are on rails: a project on your own computer that takes you from the files your community keeps to a measured model, while the people who hold the language decide where its sentences may go, what counts as good enough, and what ships. That release is out, on npm and PyPI; the rails promise you will know how good a model is — not that it is good. The releases are below.

Today's releases, on npm and PyPI — the CLI and the MCP server are source-available and free for noncommercial use; the evaluation harness is open source (AGPL):

Copies a setup prompt for your AI coding tool

Set up Champollion in this project. Champollion is machine translation development and deployment on rails with guardrails, built with low-resource and Indigenous language communities in mind.

1. Read https://champollion.dev/for-agents.md — it lists every tool and machine-readable endpoint.
2. Add the Champollion MCP server using whatever mechanism this environment supports. It runs over stdio as: npx -y champollion-mcp-server
3. Confirm by listing the MCP server's tools, then tell me which ones are available.
4. Do not open or read my data files yourself; ask me first. Asking is not a lock — for protected text, use an agent whose model runs on this computer.

Or install the tools yourself:

npm install champollionTranslation CLI

npx -y champollion-mcp-serverMCP server — for AI agents

pipx install mt-eval-harnessEvaluation harness

Contribute compute to the project queue ?The queue is a ranked list of translation benchmarks nobody has run yet. The script installs the harness, prices the work and asks before spending, then runs the top of the queue with your own API key — up to the budget you set — publishing each result to the public board.

curl -fsSL https://champollion.dev/run_queue | bash -s -- --budget 2

The site itself is in closed beta while we finish re-verifying our sources, licences, and claims.

It will open when the checking is done — and done right.

Questions, corrections, or a request for early access: info@champollion.dev. Corrections to any cited figure are especially welcome.