Local SWE Agent

Runs mini-swe-agent against a local Ollama instance (qwen2.5-coder:7b) — no cloud API keys required.

Prerequisites

• Docker (with Compose)
• uv — curl -LsSf https://astral.sh/uv/install.sh | sh

Usage

Give it a task

./run.sh "fix the bug in src/parser.py where empty strings cause a crash"

Interactive mode (no task argument)

./run.sh

How it works

run.sh does everything in order:

1. Starts the Ollama Docker container (pulls qwen2.5-coder:7b on first run — ~4 GB)
2. Waits until Ollama is ready to accept requests
3. Launches mini-swe-agent via uvx (no install needed), pointed at the local model

Configuration lives in mini.yaml. The model runs entirely on your machine.

Code Statistics

Language	Files	Lines	Blanks	Comments	Code	Complexity
Markdown	2	54	16	0	38	0
Python	2	11	2	0	9	1
Shell	2	26	5	5	16	3
YAML	2	117	22	4	91	0
Dockerfile	1	13	4	2	7	3
JSON	1	10	0	0	10	0
Total	10	231	49	11	171	7

• Estimated Cost to Develop (organic): $4,229
• Estimated Schedule Effort (organic): 1.72 months
• Estimated People Required (organic): 0.22
• Processed: 6,888 bytes (0.007 megabytes)

Generated with scc on 2026-05-22