CowAgentvsUmaDev
Generated from the two spec rows. Green marks the better value where a spec has a clear direction. Everything else is just different.
CowAgent
zhayujie · Agent harness
OSSMCP
Panel
6.44 spec wins
- Reliability
- 6.0
- Usefulness
- 6.3
- Cost
- 6.5
- Longevity
- 6.7
“The README states that the project does not issue any cryptocurrency, which tells you what three years of its inbox looked like.”
UmaDev
umacloud · Agent harness
OSS
Panel
6.41 spec wins
- Reliability
- 6.5
- Usefulness
- 6.7
- Cost
- 6.8
- Longevity
- 5.5
“It assigns product, architecture, UI, frontend, backend, QA, security and DevOps seats, which is not a CLI flag, it is a reorganisation.”
Spec by spec
| Spec | CowAgent | UmaDev |
|---|---|---|
| Architecture | ||
| Category | Agent harness | Agent harness |
| Runs | local | local |
| Platforms | macos, linux, windows | macos, linux, windows |
| Context window | not documented | not documented |
| Protocols | ||
| MCP client | Yes | No |
| MCP server | No | No |
| Capabilities | ||
| Runs terminal commands | Yes | Yes |
| Multi-file edits | No | YesCode is generated and written by the base CLI UmaDev drives; the coordinator role plans and gates rather than editing files. |
| Git operations | No | No |
| Browser control | Yes | No |
| Sandboxed execution | No | No |
| Multi-agent orchestration | Yes | Yes |
| Headless / CI mode | No | No |
| Models | ||
| Backbone | Claude, OpenAI, Gemini, DeepSeek, Qwen, GLM, Kimi, MiniMax, Doubao, ERNIE, LinkAI | Claude Code, Codex, OpenCode, Grok Build, Kimi Code |
| Bring your own model | Yes | Yes |
| Local models | Yes | NoAn optional local embedding model is used for retrieval only; the coding model always comes from the base CLI. |
| Cost | ||
| Pricing model | byok | byok |
| Starts at | $0/mo | $0/mo |
| Free tier | Yes | Yes |
| Bring your own key | Yes | Yes |
| Openness | ||
| Open source | Yes | Yes |
| License | MIT | MIT |
| GitHub stars | 47,207 | 260 |
Which one would each critic pick
| Critic | CowAgent | UmaDev | Pick |
|---|---|---|---|
| El Juez | — | — | not enough reviews |
| El Amigo | 7.0 | 6.8 | CowAgent — Pick CowAgent if you want one assistant answering in WeChat, Feishu and DingTalk from a single install; pick QwenPaw if you want a sandbox under the shell tool. |
| El Crítico | 6.0 | 6.0 | no preference |
| El Profesor | 5.8 | 7.0 | UmaDev — Roles exchange bounded artefacts and structured verdicts, and the tool reports incomplete work as incomplete rather than presenting every run as a success. |
| La Inversora | 6.5 | 5.8 | CowAgent — A rename from chatgpt-on-wechat carries 46k stars into a new brand, and the business is LinkAI, which sells hosting, role-based access and private deployment on top. |
| La Jefa | 5.5 | 5.8 | UmaDev — The coordinator leaves an audit trail, which is more than most of this category offers, and there is still no SSO and nothing tying that trail to a person. |
| El Hacker | 7.5 | 7.0 | CowAgent — MIT, a Custom provider row for local models or a proxy, one mcp.json with stdio and SSE that hot-reloads, and skills installable from GitHub or ClawHub with a slash command. |
Picks are derived from each critic's own scores. Humans vote on matchups on the duels page.