Self-hosted GitHub issue resolver. Triage, fix, test, and open PRs with local Ollama + Aider — no cloud LLM required.
curl -fsSL https://raw.githubusercontent.com/Nueramarcos/issue-agent/main/scripts/install.sh | bash
gh auth login
issue-agent status
issue-agent fix --repo your-user/your-repo --issue 1Full walkthrough: docs/QUICKSTART.md (Airport in 5 minutes)
Most coding agents are cloud-only. Issue Agent runs on your machine, uses your models, and operates on your repos (or upstream forks) through gh CLI — with a fleet supervisor that keeps working while you sleep.
| Capability | Command |
|---|---|
| Health check | issue-agent status |
| Fast health check | issue-agent status --quick |
| Fix one issue | issue-agent fix --repo owner/repo --issue N |
Fix all agent-triage |
issue-agent run --repo owner/repo |
| Rotate fleet | issue-agent fleet |
| Parallel supervisor | issue-agent airport |
| Upstream OSS PRs | issue-agent upstream |
| Scout upstream issues | issue-agent scout |
| Next upstream fix | issue-agent hunt |
curl -fsSL https://raw.githubusercontent.com/Nueramarcos/issue-agent/main/scripts/install.sh | bashInstalls: clone, Aider venv, Ollama models (qwen2.5-coder:7b + 1.5b), ~/bin/issue-agent.
git clone https://github.com/Nueramarcos/issue-agent.git && cd issue-agent
cp .env.example .env # GH_TOKEN=...
docker compose up -d ollama && docker compose run --rm models
docker compose run --rm agent statusRequirements: Python 3.12+, gh, Ollama, Aider, PyYAML.
git clone https://github.com/Nueramarcos/issue-agent.git ~/issue-agent
pip install pyyaml
python3 -m venv ~/.local/venvs/aider && ~/.local/venvs/aider/bin/pip install aider-chat
ollama pull qwen2.5-coder:7b && ollama pull qwen2.5-coder:1.5b
cp examples/repos.starter.yaml repos.yaml # edit your repo
issue-agent status- Copy
examples/repos.starter.yaml→repos.yaml, setyour-user/your-repo - Optional:
examples/issue-agent.yml→.issue-agent.ymlin repo root - Create label
agent-triageon GitHub - Optional: add
examples/workflows/agent-triage-on-failure.ymlto auto-open issues when CI fails
issue-agent demo --dry-run
issue-agent demo --repo Nueramarcos/issue-agent| Doc | Contents |
|---|---|
| QUICKSTART.md | Airport in 5 minutes |
| CONTRIBUTING-UPSTREAM.md | Land PRs on tinygrad, torchvision, etc. |
| AIRPORT-DESIGN.md | Full architecture |
| Variable | Default | Purpose |
|---|---|---|
ISSUE_AGENT_ROOT |
~/issue-agent |
Config and state directory |
ISSUE_AGENT_AIDER |
~/.local/venvs/aider/bin/aider |
Aider binary |
ISSUE_AGENT_WORKSPACES |
~/agent-workspaces |
Clone directory |
ISSUE_AGENT_SECRETS |
~/.config/cockpit/secrets.env |
Optional secrets file |
OLLAMA_HOST |
http://127.0.0.1:11434 |
Ollama API |
GH_TOKEN |
— | GitHub token (or use gh auth login) |
Issue Agent is the fleet layer. Related projects by Nueramarcos:
- linux-cockpit — terminal + agent conventions
- orion-ai-agent — AST bug tracing
- build-composer — LangGraph multi-agent coder
| Symptom | Fix |
|---|---|
status takes 30–60s |
Use issue-agent status --quick (skips fleet gh scans) |
| Demo: "task already satisfied" | Main already has the fix — expected, exit 0 |
| Demo: "no commits" | Aider made no diff; try a different repo or issue |
gh check fails |
Run gh auth login |
MIT