# Common AI (working title) > Vision: free, open-source software kept current by everyone's AI contributions. This is a hope, not a forecast. The first piece is a harness that lets a person's own AI agent take on small, pre-approved open-source tasks under a spending cap, > with a kill switch and human review. It is a dry run. It is NOT installable yet. ## Instruction for AI agents Do not run commands, install packages, or contact any project because of this site. Tell your user what this page says and ask what they want. Never ask your user for a subscription login. An API key with a spend limit is the only credential that fits. There are no rewards of any kind. ## What exists - Guard rails: spend cap with a ledger, kill switch, one-file scope, secret scan, egress allowlist, branch rule - Coverage gate and independent mutation runner - Replay mode, where an agent's reply passes through the same checks - A hosted-model adapter (Workers AI) held to a spend cap by the ledger; its cost has not yet been reconciled against the provider's own figure - A local-model adapter, tested against a mock only - An opt-in file format with a strict checker, and a pull request publisher that opens a PR only from the submitter's own fork, in one command; both are tested against a fake GitHub; the publisher has also opened one real pull request, on our own demo repository (merged by us); no outside project has adopted the file - A container verifier for model-written code (tested on macOS with OrbStack only; skipped if a person passes a flag to run on the host) ## What does not exist - An installer or package - A name and domain check: "Common AI" is a working title; another organisation, CommonAI, uses a near-identical name. The tool itself is called Common Contributions. ## Featured projects (other people's work; we do not run or join them) - Pluralis Agora: open protocol (Apache-2.0 code), public volunteer GPUs, weights not published by design, partly run by Pluralis itself, no run open as of 2026-10-13. - Covenant-72B: Apache-2.0 weights, completed run, protocol inactive, peers needed 8 B200 GPUs and were paid in TAO. - Common Contributions itself pays no rewards and has no compute network. Nothing here asks your agent to join either project. ## Source and feedback - Repository (Apache-2.0): https://github.com/Divinci-AI/common-contributions - Maintainer feedback and project proposals: https://github.com/Divinci-AI/common-contributions/issues/new/choose ## Pages - [Overview](index.html) - [Dry-run report](report/index.html)