Why open source

The method is the proof, and proof needs readers.

Every claim on this site points at a file in a public repository. That is not a distribution choice made after the fact. It is the reason the project can say what it says.

Four reasons

Why the code is open.

The approval is a pull request.

Agents propose, humans approve, and the approval is a merged pull request with a named person on it. That sentence only means something if anyone can read the pull requests, the register that explains them, and the findings that measured them. They can, all of it, from the first commit.

Reproducibility needs strangers.

A clone of main at any tag gets a working demo: make doctor, then make demo, keyless. Two machines that pull the same commit build the same gold, and the demo digest matched across a Linux sandbox, a Mac, and a Windows runner. Those are claims a stranger can falsify, which is what makes them worth making.

The parts are open, so the whole is.

ODCS is a Linux Foundation standard. dbt Core, DuckDB, PyAirbyte, and the MCP SDK are open source. A reference implementation of how they fit together would be a strange thing to close.

Apache-2.0, and nothing behind it.

There is no pricing, no hosted tier, and no service. The demo runs on your machine and the only database artifact the project publishes is the export attached to a release. The stable line changes only through the decision register, and every change is a reviewed pull request.

What open source means here

A stable line, kept honestly.

The stable line began at v1.0.0. The pipeline, the engine, the serving layer, the two proposers, and the gates change only through the decision register; experiments live on branches or under a documented experiments folder and end in written findings; a clone of main at any tag gets a working demo. That promise is written on every release surface and in the coding agent's own instructions.

It does not mean a service level. There is no support commitment, no uptime, and no performance figure, and the non-goals that keep it that way are written into the README: multi-tenancy, auth, billing, any UI beyond the read-only demo, streaming, orchestration platforms, autonomous multi-step agents, more than one connector type, and production service levels.

The one exception

The name and the logo.

The MetricMine name and the MetricMine logo are not covered by the Apache-2.0 license. Forks and derivative works may not use them in a way that suggests endorsement by, or affiliation with, this project. Everything else in the repository is yours to use under the license in LICENSE, with the provenance statement in NOTICE beside it.

The website and the brand system live in their own repository: MIT for the site's code, all rights reserved for the brand assets, and the normative brand standard beside them.

Taking part

Small reviewed pull requests, contracts never weakened.

Contributions are welcome under the conventions in CONTRIBUTING.md: small reviewed pull requests, a contract never weakened to pass a gate, a version bump on every contract change, and a sign-off on every commit, which the DCO check enforces. Everyone taking part follows the code of conduct. Vulnerabilities go through SECURITY.md, never a public issue.

The issue forms know the project's shapes: a bug, a contract change, and a finding. A finding is the project's favorite kind of issue, because a measurement that contradicts the plan is how the plan improves, and sixty-six of them are on record already.

Next

Read it, run it, or file a finding.