SolonCodevsTianshu Harness
Generated from the two spec rows. Green marks the better value where a spec has a clear direction. Everything else is just different.
SolonCode
OpenSolon · Terminal agent
#77OSSMCP
Panel
6.02 spec wins
- Reliability
- 5.5
- Usefulness
- 5.8
- Cost
- 7.0
- Longevity
- 5.8
“The command line ships pointed at a cheap default model, which is the most realistic assumption any of these tools has made about its users.”
Tianshu Harness
Tianshu · Terminal agent
#145OSSMCP
Panel
6.00 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 | SolonCode | Tianshu Harness |
|---|---|---|
| Architecture | ||
| Category | Terminal agent | Terminal agent |
| Runs | local | local |
| Platforms | macos, linux, windows, web | 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 | YesThe desktop workspace shows Git changes alongside the editor and terminal. | Yes |
| Browser control | Yessoloncode web serves the agent UI over HTTP; the README does not describe a browser-driving tool. | No |
| Sandboxed execution | No | NoSandboxing is a documented permission model over the agent's own tools, not a container. |
| Multi-agent orchestration | YesAgents are one of the extensible runtime types and are addressed with @ in the CLI; the README's own worked example asks it to design an agent team. | No |
| Headless / CI mode | No | No |
| Models | ||
| Backbone | deepseek-v4-flash, any configured provider | DeepSeek V4, GLM, OpenAI, Ollama |
| Bring your own model | Yes | Yes |
| Local models | YesModel providers are configured freely on the Settings page, so a local endpoint works. | 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 | n/a | 1,013 |
Which one would each critic pick
| Critic | SolonCode | Tianshu Harness | Pick |
|---|---|---|---|
| El Juez | — | — | not enough reviews |
| El Amigo | 6.5 | 6.3 | SolonCode — Pick it if you want an agent you can rewind after it goes wrong; pick a diff-review tool if you would rather it never got that far. |
| El Crítico | 5.0 | 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 | 6.0 | 5.5 | SolonCode — Model, token and time statistics are reported per response, which is measurement built into the interaction rather than an estimate produced afterwards. |
| La Inversora | 6.0 | 5.5 | SolonCode — A framework team shipping an agent, published on a domestic code host first with a mirror abroad, and released in February 2026 with no adoption recorded yet. |
| La Jefa | 5.3 | 5.3 | no preference |
| El Hacker | 7.5 | 7.8 | Tianshu Harness — Apache-2.0, tool servers attach over the protocol, and a local runtime sits on the provider list beside the hosted ones, so nothing has to leave the machine if I do not want it to. |
Picks are derived from each critic's own scores. Humans vote on matchups on the duels page.