Atomic AgentvsTianshu Harness
Generated from the two spec rows. Green marks the better value where a spec has a clear direction. Everything else is just different.
Atomic Agent
AtomicBot · Terminal agent
#187OSSMCP
Panel
6.02 spec wins
- Reliability
- 5.2
- Usefulness
- 6.0
- Cost
- 8.7
- Longevity
- 4.3
“It keeps a browser profile as a local file, so your agent now has browsing history and you can read it.”
Tianshu Harness
Tianshu · Terminal agent
#145OSSMCP
Panel
6.01 spec wins
- Reliability
- 5.5
- Usefulness
- 6.0
- Cost
- 7.5
- Longevity
- 5.0
“Its release badges point at a different repository than the one you cloned, which is a versioning strategy of sorts.”
Spec by spec
| Spec | Atomic Agent | Tianshu Harness |
|---|---|---|
| Architecture | ||
| Category | Terminal agent | Terminal agent |
| Runs | local | local |
| Platforms | macos, linux, windows | macos, linux, windows |
| Context window | not documented | not documented |
| Protocols | ||
| MCP client | Yes | Yes |
| MCP server | No | No |
| Capabilities | ||
| Runs terminal commands | Yes | Yes |
| Multi-file edits | Yes | Yes |
| Git operations | No | Yes |
| Browser control | Yes | No |
| Sandboxed execution | No | NoSandboxing is a documented permission model over the agent's own tools, not a container. |
| Multi-agent orchestration | No | No |
| Headless / CI mode | No | No |
| Models | ||
| Backbone | llama.cpp, any OpenAI-compatible endpoint, Anthropic, OpenAI | DeepSeek V4, GLM, OpenAI, Ollama |
| Bring your own model | Yes | Yes |
| Local models | Yes | Yes |
| 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 | Apache-2.0 |
| GitHub stars | 2,564 | 1,013 |
Which one would each critic pick
| Critic | Atomic Agent | Tianshu Harness | Pick |
|---|---|---|---|
| El Juez | — | — | not enough reviews |
| El Amigo | 6.3 | 6.3 | no preference |
| El Crítico | 5.5 | 5.8 | Tianshu Harness — What the documentation calls a sandbox is a permission model over the agent's own tools, not a container, so nothing constrains a process once the agent has been allowed to start it. |
| El Profesor | 5.5 | 5.5 | no preference |
| La Inversora | 5.5 | 5.5 | no preference |
| La Jefa | 5.5 | 5.3 | Atomic Agent — Nothing per seat and nothing leaves the workstation, which is the shortest data review I will run, and there is no console over sixty of them. |
| El Hacker | 8.0 | 7.8 | Atomic Agent — MIT, a llama.cpp build underneath, MCP servers wired in, and it embeds over HTTP or as a Tauri sidecar, so it becomes a component in things I build. |
Picks are derived from each critic's own scores. Humans vote on matchups on the duels page.