Run mini-SWE-agent in Agentastic
The deliberately small SWE-bench agent, useful for focused tasks and reproducible experiments.
What Agentastic adds
Give every task its own worktree, branch, or container.
Keep mini-SWE-agent's real CLI and your existing configuration.
Inspect the diff, run code review, test, and open a pull request.
Launch mini-SWE-agent
- Install the official CLI and sign in with your existing provider account.
- Open a repository in Agentastic. The connection is discovered from your PATH.
- Open Agent Home, choose mini-SWE-agent, describe the task, and select worktree, local, container, cloud, or remote mode.
Every run gets its own git worktree, so three tasks can share a repository without sharing a branch or a build directory.
What mini-SWE-agent supports
- Launch from a task description: supported
Type the task in Agent Home and Agentastic hands it to mini at launch, so the run starts from your prompt instead of an empty session.
- Unattended runs: supported
mini-SWE-agent can run without stopping for per-tool confirmation, which is what makes an overnight or scheduled run finish on its own.
- Plan before editing: not supported
Agentastic keeps plan mode out of the picker for mini-SWE-agent rather than emulating it, so runs begin with edits.
- Resume a session: not supported
Every launch of mini-SWE-agent is a fresh session; carry context forward in the repository rather than in the agent.
- Structured chat: not supported
mini-SWE-agent runs in its own terminal UI inside Agentastic; it does not expose a protocol Agentastic can render as a structured chat.
- Code review: not supported
mini-SWE-agent writes the change; pair it with a review-capable agent for the pass before merge.
Agentastic reads these from mini-SWE-agent's own CLI rather than emulating them, and modes it does not support stay out of the picker. The provider reference lists the exact flags for each launch, and how to compare coding agents covers what changes when you switch harnesses.
mini-SWE-agent questions
Is there a GUI for mini-SWE-agent?
Yes. Agentastic runs the real mini-SWE-agent CLI (mini) in a native Mac workspace with a git worktree, terminal, diff view, and code review for each task.
How do I run multiple mini-SWE-agent sessions in parallel?
Start each task from Agent Home in worktree mode. Agentastic gives every run its own git worktree and terminal, so several mini-SWE-agent sessions can work in one repository without editing the same files.
How do I install mini-SWE-agent?
Run uv tool install mini-swe-agent, then sign in the way mini-SWE-agent's documentation describes. Agentastic finds the mini command on your PATH when you open a repository.
Does Agentastic charge for mini-SWE-agent?
No. Agentastic is free and does not proxy model calls. It launches mini-SWE-agent with the credentials and plan you already use, so any usage is billed by your provider, not by Agentastic.