mini-SWE-agentvsphi
Generated from the two spec rows. Green marks the better value where a spec has a clear direction. Everything else is just different.
mini-SWE-agent
SWE-agent (Princeton and Stanford) · Terminal agent
#35OSS
Panel
6.93 spec wins
- Reliability
- 6.3
- Usefulness
- 5.5
- Cost
- 8.8
- Longevity
- 7.0
“One hundred lines of Python, which is fewer than most competitors spend on their pricing page.”
phi
Pulse AI Club · Terminal agent
#28OSSMCP
Panel
6.92 spec wins
- Reliability
- 7.2
- Usefulness
- 6.5
- Cost
- 8.0
- Longevity
- 6.0
“Its extensions talk over a binary protocol on standard input and output, because JSON was apparently the slow part of asking a model.”
Spec by spec
| Spec | mini-SWE-agent | phi |
|---|---|---|
| Architecture | ||
| Category | Terminal agent | Terminal agent |
| Runs | local, sandbox | local |
| Platforms | macos, linux | macos, linux, windows |
| Context window | not documented | not documented |
| Protocols | ||
| MCP client | No | Yes |
| MCP server | No | No |
| Capabilities | ||
| Runs terminal commands | Yes | Yes |
| Multi-file edits | Yes | Yes |
| Git operations | Yes | No |
| Browser control | No | NoExternal HTTP fetch is available only through an MCP server you configure. |
| Sandboxed execution | Yes | No |
| Multi-agent orchestration | No | Yes |
| Headless / CI mode | Yes | Yes"phi run -p" executes one agent loop headlessly, and a headless-strict mode is one of the configurable permission modes. |
| Models | ||
| Backbone | any via litellm | OpenAI-compatible, Anthropic |
| Bring your own model | Yes | Yes |
| Local models | Yes | YesAny OpenAI-compatible base URL can be configured, including a locally served endpoint. |
| Cost | ||
| Pricing model | byok | byok |
| Starts at | n/a | $0/mo |
| Free tier | Yes | Yes |
| Bring your own key | Yes | Yes |
| Openness | ||
| Open source | Yes | Yes |
| License | MIT | MIT |
| GitHub stars | 8,145 | 524 |
Which one would each critic pick
| Critic | mini-SWE-agent | phi | Pick |
|---|---|---|---|
| El Juez | — | — | not enough reviews |
| El Amigo | 6.8 | 7.3 | phi — Pick phi if you have ever had an agent quietly mangle a file; pick a more forgiving terminal agent if you would rather it guessed than stopped. |
| El Crítico | 6.5 | 6.3 | mini-SWE-agent — There is no procedure for anything beyond a shell command, opening a pull request is tell the LM to figure it out, and without a sandbox flag every command lands on the host. |
| El Profesor | 8.0 | 7.5 | mini-SWE-agent — The reference harness for the SWE-bench bash-only leaderboard: one bash action per turn via subprocess.run, a linear history, no tool-calling API, and Gemini 3 Pro reported above 74% on Verified. |
| La Inversora | 6.8 | 5.5 | mini-SWE-agent — No company, a Princeton and Stanford lab with 6,938 stars; the funding is grants and the exit is a paper, which is more stable than half the cap tables on this board. |
| La Jefa | 5.0 | 6.3 | phi — Free at any headcount, and it runs one loop non-interactively with a strict headless permission mode, which makes it the rare desktop tool that can also be a pipeline step. |
| El Hacker | 8.5 | 8.8 | phi — MIT, any compatible base URL including one I serve myself, and tool servers listed by name only until the agent calls list, inspect and invoke to reach them on demand. |
Picks are derived from each critic's own scores. Humans vote on matchups on the duels page.