Tencent open-sources TeamAI-CLI to put every coding agent on one Git leash

Why it matters

Coding agents are multiplying faster than teams can govern them. Tencent is using Git to make agent instructions portable, reviewable and independent of any one vendor.

Tencent has open-sourced TeamAI-CLI, an internal tool that uses Git to distribute a team's rules, skills and accumulated knowledge across multiple AI coding agents.

Tencent AI (@TencentAI_News) said in an X post on September 7th that Tencent had used TeamAI-CLI internally since March. The public TeamAI-CLI repository is available under the MIT license, which permits commercial use, modification and redistribution.

The release addresses a problem created by the rapid adoption of AI coding tools inside engineering organizations: each agent has its own directories, configuration files, skills and instruction formats. Teams can end up maintaining separate versions of the same operating rules for Claude Code, Codex, Cursor, Tencent's CodeBuddy and other agents.

TeamAI-CLI turns a shared Git repository into the control plane for those instructions. Skills, rules, documentation, hooks, environment declarations and Model Context Protocol server configurations can live in the repository. Changes move through the familiar branch, review and merge process before the CLI distributes approved files into each supported agent's local directories.

That design gives agent instructions some of the controls engineering teams already apply to source code. A proposed rule can be reviewed before reaching every developer. Git preserves its history. A bad change can be traced or reverted. The repository also creates a single place to inspect what an organization's agents have been told to do, rather than leaving behavior scattered across employee laptops and vendor-specific configuration folders.

Git becomes the agent management layer

Administrators create a repository and run teamai init against it. Team members initialize TeamAI-CLI inside a project or at the user level. At the beginning of an AI session, a hook can pull the latest approved resources and install them into the relevant agent directories.

Contributors use teamai push to place local changes on a branch and open a merge request. Once reviewers approve and merge the change, subsequent pulls distribute it across the team. TeamAI-CLI also supports role and tag controls, allowing organizations to deliver different instructions to different groups without maintaining a separate system for every agent vendor.

The public repository lists support across Claude Code, Codex, Cursor, CodeBuddy, WorkBuddy, OpenCode, Qoder and several other agent environments. Git hosting can sit on GitHub, GitLab, Tencent's TGit, CNB or a private Git service. The package is installed through npm and exposes a teamai command.

Tencent's choice of Git is deliberately conservative. Engineering organizations already have access controls, review conventions and audit trails built around repositories. TeamAI-CLI extends those mechanisms to the prompts, skills and operational knowledge that increasingly determine how coding agents modify a codebase.

The approach also limits dependence on any single coding agent. A team can keep its operating instructions in a repository it controls, then translate and distribute those resources into several tools. That portability matters as developers switch agents or use different products for different projects.

Tencent is also packaging institutional memory

TeamAI-CLI extends beyond configuration synchronization. Its knowledge system can index team documentation, rules, shared lessons and a generated graph of codebase components. An optional recall agent searches that material before a task and returns relevant context to the coding agent.

The repository describes a retrieval system that combines text ranking with information from the code graph. Teams can import individual repositories or groups of repositories, producing structured records of components, interfaces, configurations and relationships between files. Search results can point agents back to source paths, reducing the amount of code exploration repeated in each session.

Tencent has also built a workflow for capturing lessons from difficult sessions. A stop hook looks for friction signals such as user corrections, denied tool calls and repeated failures. When the score crosses a threshold, TeamAI-CLI can suggest summarizing the lesson and contributing it to the shared repository. The proposal still passes through Git, preserving human review before an agent's mistake becomes organization-wide guidance.

That loop is the more consequential part of Tencent's release. Static instruction files quickly become stale as codebases, tools and models change. TeamAI-CLI is designed to turn corrections made during daily work into reviewed material that future agent sessions can retrieve.

The system also includes session summaries, usage digests and a dashboard for examining coding-agent activity. Those capabilities push TeamAI-CLI toward an internal operations layer for agent deployment, rather than a simple configuration copier. They also place responsibility on adopting teams to decide what session information should be retained and which repositories may hold sensitive instructions or technical knowledge.

Tencent's public repository had accumulated more than 800 GitHub stars, roughly 80 forks and more than 600 commits when checked on September 7th. The commit history and changelog show sustained development following the internal versions documented in March, with later additions covering knowledge retrieval, codebase graphs, MCP distribution and support for more agents.

Open-sourcing TeamAI-CLI gives Tencent outside contributors and a broader compatibility test bed while promoting Git as the neutral layer between teams and a growing list of coding-agent vendors. The bet is straightforward: models and interfaces will keep changing, while engineering organizations will still need one reviewed record of how their agents are supposed to work.