Your AI agent, right inside GitHub.
Create an issue. Get results. Artifacts committed automatically.
Get Started · What Can It Do? · How It Works · Providers · Scheduling · Storage · Memory · Compare · Questions
flowchart LR
A(Use This Template) --> B(Add API Key)
B --> C(Create Issue)
C --> D(Get Results)
style A fill:#2d333b,stroke:#539bf5,color:#adbac7
style B fill:#2d333b,stroke:#539bf5,color:#adbac7
style C fill:#2d333b,stroke:#539bf5,color:#adbac7
style D fill:#2d333b,stroke:#347d39,color:#57ab5a,stroke-width:2px
- Use this template — Click the green Use this template button above → Create a new repository → set visibility to Private.
- Add your API key — In your new repo, go to Settings → Secrets → Actions → add
OPENROUTER_API_KEY(get one here). - Create an issue — Pick any issue template. The
gitgentlabel triggers the agent automatically.
That's it. The agent picks up the issue, executes the task, posts results as comments, and commits output files to your repo.
Always create a private repo. The agent posts results on issues and commits files — a public repo would expose your data. Use "Use this template" (not Fork) so you can set it to Private.
Pick an issue template and the agent handles the rest:
| Template | What It Does |
|---|---|
| 🤖 Agent Task | General-purpose task — ask it anything |
| 📚 Research | Deep web research with cited sources |
| 📈 Data Analysis | Analyze data, generate reports and charts |
| 📊 Marketing | Competitor analysis and market research |
| ✍️ Content Writing | Articles, newsletters, documentation |
| 📄 Document Generation | Excel, Word, and PowerPoint files |
| 📰 News Briefing | News monitoring and recurring digests |
| 🔍 Job Search | Job board parsing and tracking |
| 🌐 Website Builder | Static sites with unique URLs via GitHub Pages |
| 🔄 Scheduled Task | Recurring tasks on a daily, weekly, or custom schedule |
Need to follow up? Comment /gitgent <your request> on any open issue.
%%{init: {'themeVariables': {'actorBkg': '#2d333b', 'actorBorder': '#539bf5', 'actorTextColor': '#adbac7'}}}%%
sequenceDiagram
actor User
participant GH as GitHub Issues
participant WF as GitHub Actions
participant Agent as Agent + LLM
participant FS as Repository
User->>GH: Create issue with "gitgent" label
GH->>WF: Trigger workflow
WF->>GH: React 👀, add "in-progress" label
WF->>Agent: Launch agent with prompt + tools
activate Agent
loop Execution
Agent->>Agent: Search, browse, analyze
Agent->>FS: Write artifacts to artifacts/issue-N/
end
Agent-->>WF: Done
deactivate Agent
WF->>GH: Post results + artifact links
WF->>GH: Label "completed", close issue
WF->>FS: git commit + push artifacts
- Isolated runs — each issue gets its own
artifacts/issue-<number>/directory; nothing leaks between tasks. - Label tracking — progress is tracked via labels:
in-progress→completed,needs-review, orfailed. - Automatic recovery — if an artifact-required run produces no files, a single recovery pass is attempted.
- Memory across runs — the agent reads past summaries and daily logs before starting, so it can build on prior work and avoid repeating itself. After each run it writes new summaries back to
memory/. - Concurrency control — runs are serialized per issue via GitHub Actions concurrency groups.
Works out of the box with OpenRouter (default). Want to use a different provider? Just change two settings:
| Provider | Set GITGENT_PROVIDER to |
Add this secret | Default Model |
|---|---|---|---|
| OpenRouter | openrouter (default) |
OPENROUTER_API_KEY |
minimax/minimax-m2.7 |
| OpenAI | openai |
OPENAI_API_KEY |
gpt-4.1 |
| Anthropic | anthropic |
ANTHROPIC_API_KEY |
claude-sonnet-4-20250514 |
google |
GEMINI_API_KEY |
gemini-2.5-flash |
|
| xAI | xai |
XAI_API_KEY |
grok-3 |
| Mistral | mistral |
MISTRAL_API_KEY |
mistral-large-latest |
| Groq | groq |
GROQ_API_KEY |
llama-3.3-70b-versatile |
Set GITGENT_PROVIDER in Settings → Variables → Actions and add the API key in Settings → Secrets → Actions.
More providers
| Provider | GITGENT_PROVIDER |
Secret |
|---|---|---|
| Azure OpenAI | azure-openai-responses |
AZURE_OPENAI_API_KEY |
| Cerebras | cerebras |
CEREBRAS_API_KEY |
| Hugging Face | huggingface |
HF_TOKEN |
| Kimi | kimi-coding |
KIMI_API_KEY |
| MiniMax | minimax |
MINIMAX_API_KEY |
| Vercel AI Gateway | vercel-ai-gateway |
AI_GATEWAY_API_KEY |
When using a non-OpenRouter provider, update your workflow env: block to pass the matching secret.
Add a schedule label to any issue with the gitgent label to run it on a recurring basis.
| Label | Frequency |
|---|---|
schedule:hourly |
Every hour |
schedule:daily |
Daily at 9 AM UTC |
schedule:weekly |
Mondays at 9 AM UTC |
schedule:custom |
Custom cron — add cron: 0 14 * * 1-5 to the issue body |
A scheduler workflow runs every hour, scans open issues for schedule labels, and triggers the agent automatically. Keep the issue open to continue the schedule.
GitHub Actions minutes — Private repos get 2,000 free minutes/month (Free plan). Run duration varies by task — a simple research query may finish in 1–2 minutes while complex document generation can use the full 30-minute timeout. Monitor usage in Settings → Billing → Actions and adjust schedules accordingly. Pricing details.
| Variable | Default | Description |
|---|---|---|
GITGENT_PROVIDER |
openrouter |
LLM provider id |
GITGENT_MODEL |
minimax/minimax-m2.7 |
Model id (provider-specific) |
MAX_RUNTIME_MINUTES |
30 |
Execution timeout in minutes |
Set these as repository variables via Settings → Variables → Actions.
Every run produces files — research reports, spreadsheets, presentations, websites, CSV data — committed directly to your repo under artifacts/issue-<number>/. A few things to know:
- File types — the agent can generate Markdown, CSV, Excel (
.xlsx), Word (.docx), PowerPoint (.pptx), HTML, and more. All files are committed to your repo with full git history. - Size cap — individual artifact files are limited to 25 MB. Most generated reports are well under 1 MB.
- Repo growth — artifacts accumulate over time. If your repo grows large, you can delete old
artifacts/issue-*/directories and force-push, or usegit filter-repoto clean history. - GitHub Pages — use the 🌐 Website Builder template to generate static sites. Each task gets its own URL under
docs/issue-<number>/, with an auto-generated landing page at the root linking to all sub-sites.
The agent maintains memory in memory/ so it can learn from previous runs:
| File | Purpose |
|---|---|
soul.md |
Agent identity and behavioral guidelines — protected, cannot be deleted by the agent |
preferences.yaml |
Your output and workflow preferences — customize to control default behavior |
summaries/ |
Timestamped run summaries — the agent reads these to build on past work |
daily/ |
Append-only daily logs — tracks every run for the day |
At the start of each run, the agent reads soul.md, preferences.yaml, and searches past summaries for relevant context. After finishing, it writes a new summary and daily log entry. This means:
- Recurring tasks (news briefings, scheduled market scans) automatically avoid repeating coverage from prior runs.
- Follow-up tasks on the same topic can reference what was already found.
- Preferences like default tone, output formats, and research depth are respected without repeating them in every issue.
Edit memory/preferences.yaml to change defaults. Memory files are committed alongside artifacts and grow over time — the total directory is capped at 10 MB.
When you create a repo with Use this template, it starts as an independent copy — there's no automatic link back to the original. To pull in new features, bug fixes, and workflow improvements:
cd your-gitgent-repo
git remote add upstream https://github.com/supercheck-io/gitgent.git
git fetch upstreamgit fetch upstream
git merge upstream/main --allow-unrelated-historiesBoth are open-source MIT-licensed AI tools — they solve different problems and work well together.
| Gitgent | OpenClaw | |
|---|---|---|
| What | AI agent inside GitHub | Personal AI assistant gateway |
| Interface | GitHub Issues — familiar to any developer | 23+ messaging channels (WhatsApp, Slack, Discord, etc.) |
| Mobile access | GitHub mobile app — create tasks, view results, follow up from your phone | Via the messaging channel's app (WhatsApp, Telegram, etc.) |
| Notifications | GitHub email + mobile push — automatic on every agent comment and issue close, no setup | Messaging channel's native notifications (push alerts in WhatsApp, Telegram, etc.) |
| Setup | Click "Use this template" → add one API key | Install CLI, configure gateway, self-host |
| Runs on | GitHub Actions — no server to manage | Your machine or VPS |
| Infra cost | None — uses GitHub's free tier | You provision and maintain the host |
| Security model | Runs in ephemeral GitHub-hosted runners | Runs on your own infrastructure |
| AI providers | 7+ via PI SDK | 30+ providers |
| Web research | Search + Playwright browser | Search + Playwright browser |
| Voice / media | — | Wake word, talk mode, media understanding |
| Scheduling | Label-driven (hourly / daily / weekly / custom cron) | Cron-based |
| Extensibility | Skills (Markdown files) | 80+ plugins |
| Output storage | Artifacts committed to your repo with full git history | Chat replies in messaging channels |
| GitHub integration | Native — issues, PRs, labels, comments | Via plugin |
| Best for | Automated research, reports, documents, batch tasks | Real-time conversations across messaging apps |
Gitgent handles unattended research and document tasks right inside GitHub — zero infrastructure, nothing to host. OpenClaw gives you a real-time AI assistant across WhatsApp, Slack, Discord, and more.
- Bug reports & feature requests → GitHub Discussions
- Contributing → CONTRIBUTING.md · Security
- Code of conduct → CODE_OF_CONDUCT.md
- License → MIT — free and open source, always.
Open source by Supercheck