Tencent Cloud has open-sourced TeamAI, a command-line tool that lets a team keep its AI coding agents’ shared setup in a Git repository, according to China AI Wire on 10 October 2026 [1]. The code is on GitHub under an MIT licence [2][3]. Tencent’s test figures are its own claims, covered below.
Tencent Cloud’s TeamAI keeps agent setup in one Git repository
An agent “harness” is everything wrapped around an AI model to make it behave the way a team wants: instructions, tools and settings. TeamAI keeps those in one shared Git repository. The README of an earlier version called it “The team harness for AI agents” and described “Git-native management of skills, rules, MCP, env vars, knowledge base, and more” across agents such as Claude Code, Codex, CodeBuddy, WorkBuddy and OpenCode [4]. The current README headline is “Make Every Team AI Native” [2].
In plain terms, skills are reusable instruction packs for a task, the same idea Anthropic has been pushing as an open format (it gave Agent Skills to the Agentic AI Foundation). Rules are standing instructions. MCP, the Model Context Protocol, is a standard way to plug tools into an agent (see our MCP explainer). Env vars are shared settings; the project’s docs say not to put secret values there [5].
For another look at harnesses, see our piece on whether agents need elaborate ones. The flow, in the project’s words, is “push → review & merge → pull”. A member runs teamai push, which opens a merge request; once a reviewer merges it, teamai pull runs at the start of a session and syncs the update to each person’s tools [5].
Setting up TeamAI takes an npm install and a shared repository
From the README: install with npm install -g teamai-cli, create a repository on a Git host (GitHub, GitLab, GitCode, CNB, TGit or a private service), give members write access, then run teamai init with its address [2]. The getting-started guide adds Node.js 20 or newer and Git [6]. The README’s table lists 19 agent entries, including regional variants; the news reports say 16 [1][2].
TeamAI is MIT-licensed and has been public since April
The LICENSE file says “teamai-cli is licensed under MIT” [3]. GitHub’s API shows the repository was created on 27 April 2026 [7], so 10 October is the announcement, not the start of the project.
Update, 1 November 2026: The repository is still MIT-licensed and active. GitHub lists version 0.28.1, published on 28 October, and shows about 5,700 stars and nearly 490 forks [7][10]. Stars measure interest, not use inside companies.
The test figures: reported, Tencent’s own
China AI Wire reports Tencent Cloud’s claim: “in an internal test involving 13 development tasks, integrating shared team knowledge into lower-cost models improved comprehensive quality scores by 6.1% while reducing model invocation costs by 76%” [1]. Chinese outlets CNMO and Guandian describe the comparison as a cheaper model with team knowledge against a higher-spec model without it [8][9]. China AI Wire also says TeamAI is integrated into Tencent’s ClawPro agent platform [1].
Tencent’s 76% saving rests on one unpublished internal test
The interesting idea in TeamAI is not the code but where it puts agent instructions: in version control, behind a merge request. That treats a prompt or a skill like any other piece of team software, with an author, a reviewer and a history that can be rolled back. For companies worried about staff quietly feeding agents inconsistent or unsafe instructions, a reviewed, shared rulebook is a sensible control, and it fits a wider push to make coding agents checkable, from Microsoft’s Scope evaluation platform to Chainguard’s tools for containing agents.
The headline numbers deserve far less weight. The 6.1% quality gain and 76% cost cut come from Tencent Cloud itself, from 13 development tasks, and TSN found no published method, dataset or scoring rubric. Comparing a cheaper model plus team knowledge against a pricier model without it also mixes two changes at once, so the test cannot show how much of the saving comes from TeamAI. Nothing published shows how many teams use the tool, and TSN found no Tencent page confirming the ClawPro integration. The same review discipline TeamAI asks for applies to shared settings too: the project’s own docs warn against putting secrets in them [5].
Related on TSN: MCP Explained: The USB-C for AI Tools (Model Context Protocol Guide); Do AI Agents Need Elaborate Harnesses? A Minimal-Agent Study and Two New Toolkits
Sources
- China AI Wire, “Tencent Cloud Open-Sources TeamAI Agent Collaboration Tool With Git-Based Skill Sharing”, 10 October 2026 (labels its reporting AI-assisted; cites Tencent Cloud). https://chinaaiwire.com/news/tencent-cloud-teamai-open-source-agent-collaboration-20261010
- Tencent, teamai-cli repository and README, GitHub (checked 10 October 2026). https://github.com/Tencent/teamai-cli
- Tencent, LICENSE, teamai-cli. https://github.com/Tencent/teamai-cli/blob/main/LICENSE
- Tencent, earlier README revision (commit of 25 August 2026). https://github.com/Tencent/teamai-cli/blob/12a49d5898c6c786de32a07b533797f734549977/README.md
- Tencent, product overview, teamai-cli docs. https://github.com/Tencent/teamai-cli/blob/main/docs/product-overview.md
- Tencent, Getting Started, teamai-cli docs. https://github.com/Tencent/teamai-cli/blob/main/docs/guide/getting-started.md
- GitHub REST API, repository record (created 27 April 2026). https://api.github.com/repos/Tencent/teamai-cli
- CNMO, “腾讯云开源TeamAI!适配WorkBuddy等十六种Agent”, 10 October 2026 (in Chinese). https://m.cnmo.com/news/819988.html
- 观点网 (Guandian), “腾讯云开源TeamAI协作工具 适配16种Agent 内测调用成本降76%”, 10 October 2026 (in Chinese). https://www.guandian.cn/m/show/611329
- GitHub, teamai-cli releases (v0.28.1, 28 October 2026; checked 1 November 2026). https://github.com/Tencent/teamai-cli/releases
- Model Context Protocol specification. https://modelcontextprotocol.io/specification/2025-06-18
Follow @tsnmediahub on X for breaking AI news as it happens.
Follow on X





