TeamAI CLI

Why it's ranked here

The verdict here is positive on engineering depth, cautious on maturity. This is not a thin dotfile-sync script: it ships a real git-provider abstraction with working implementations for six hosts, including a from-scratch GitLab REST client for self-hosted instances and a deliberately limited generic-git fallback that refuses to fake operations it cannot perform. Underneath the sync layer sits a second subsystem, a source-code knowledge graph built on real AST parsing, with confidence scoring, staleness detection and pruning for the learnings it captures. That is substantially more machinery than the pitch of syncing skill files implies. Two of the product's three stated layers are still labeled beta by the project itself.

What's good

The provider layer is where the engineering shows. GitLab support talks the REST API directly rather than shelling out to a CLI, so it works against self-hosted enterprise instances behind their own URL and API prefix, not just the public host. Credential handling is careful: a self-hosted GitLab probe times out in three seconds, never sends a token and never follows a redirect before asking a human to confirm the instance, and provider tokens are written to disk with restricted permissions rather than embedded in a stored URL. Machine-local state such as the search index and provider clones lives outside the project's own repository, keeping the business repo untouched.

Tradeoffs

The generic git backend, the fallback for any host the other five providers don't recognize, is transport-only: it cannot create a repository or open a pull request, and calling either throws an explicit error rather than pretending to succeed. That pushes real admin work onto a human for teams on a plain self-hosted Git server. Two providers depend on installing a separate platform CLI first, one for Tencent's internal host and one for cnb.cool, an external dependency the GitHub and GitLab paths avoid. The tool also migrates its own on-disk layout between versions with no supported way back: downgrading afterward is not supported, per its own documentation.

How to use it well

This fits a team already on git review that runs two or more AI coding tools daily, since the payoff is the setup work it removes across those tools. An admin starts by creating the shared repository, then chooses project scope for repo-specific skills or user scope for anything wanted everywhere. Teams on GitHub or self-hosted GitLab get the smoothest path, since those providers open review requests automatically; teams on a plain self-hosted git server open every request by hand instead. It is not a fit for a solo developer, or a team standardized on one AI tool, since nothing needs keeping in sync.

Observed

License

MIT

Install / interfaces

npm global install; single teamai command-line binary, no GUI or hosted API

Git provider support

Six backends implemented: GitHub, GitLab (including self-hosted), GitCode, CNB, TGit, and a generic git fallback

Platform requirement

Node.js 20 or later; Git required on PATH

Test tooling

Vitest-based unit, coverage and end-to-end test scripts declared in the package manifest

Knowledge engine dependency

Ships AST parsing (tree-sitter) for its code-knowledge-graph subsystem, not just file copying

Data isolation

Machine-local state (config, clones, search index) is stored outside the project repository, under a per-project partition in the user's home directory

Product maturity labeling

Two of the three stated product layers (Team Context, Team Improvement) are marked beta in the project's own documentation