mini-SWE-agentvsPoolside
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
#8OSS
Panel
6.91 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.”
Poolside
poolside · Terminal agent
#15MCP
Panel
6.92 spec wins
- Reliability
- 6.8
- Usefulness
- 7.0
- Cost
- 6.5
- Longevity
- 7.2
“One of its three network policies is named unsafe-allow-all, which is at least the most clearly labelled decision on this board.”
Spec by spec
| Spec | mini-SWE-agent | Poolside |
|---|---|---|
| Architecture | ||
| Category | Terminal agent | Terminal agent |
| Runs | local, sandbox | local, sandbox, cloud |
| Platforms | macos, linux | macos, linux, windows |
| Context window | not documented | 1M tokens on Laguna S 2.1, 256k on XS 2.1 and M.1 |
| Protocols | ||
| MCP client | No | Yes |
| MCP server | No | No |
| Capabilities | ||
| Runs terminal commands | Yes | Yes |
| Multi-file edits | Yes | Yes |
| Git operations | Yes | Yes |
| Browser control | No | NoWeb search and fetch are documented; there is no DOM-level browser automation. |
| Sandboxed execution | Yes | YesTool commands run inside a container on your machine and need a local Docker engine; network policy is off, allow-list or unsafe-allow-all, enforced through a proxy container. |
| Multi-agent orchestration | No | Yes |
| Headless / CI mode | Yes | Yes`pool exec --prompt ... --output json` with POOLSIDE_API_KEY, plus a documented GitHub Actions integration. |
| Models | ||
| Backbone | any via litellm | Laguna S 2.1, Laguna XS 2.1, Laguna M.1 |
| Bring your own model | Yes | Yes |
| Local models | Yes | YesDocumented local-run guides for Ollama and vLLM on Metal, plus full on-premises and air-gapped deployment. |
| Cost | ||
| Pricing model | byok | mixed |
| Starts at | n/a | n/a |
| Free tier | Yes | Yes |
| Bring your own key | Yes | Yes |
| Openness | ||
| Open source | Yes | No |
| License | MIT | proprietary |
| GitHub stars | 8,410 | n/a |
Which one would each critic pick
| Critic | mini-SWE-agent | Poolside | Pick |
|---|---|---|---|
| El Juez | — | — | not enough reviews |
| El Amigo | 6.8 | 6.8 | no preference |
| El Crítico | 6.5 | 6.5 | no preference |
| El Profesor | 8.0 | 6.5 | 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 | 7.0 | Poolside — Training your own models and selling deployments into other people's data centres is a capital-intensive bet on the one segment that cannot buy from a lab directly. |
| La Jefa | 5.0 | 7.0 | Poolside — It runs inside our own network, including air-gapped, on the Kubernetes platforms we already operate, and the enterprise price is a sales conversation. |
| El Hacker | 8.5 | 7.5 | mini-SWE-agent — MIT, any model through litellm, OpenRouter or Portkey including my local server, a YAML config, and source short enough to read before breakfast; the missing MCP client is the only gap. |
Picks are derived from each critic's own scores. Humans vote on matchups on the duels page.