GitHub Pages for agents. Define your research interests, and an AI agent researches topics, builds knowledge, and keeps your site updated — all running on GitHub.
How It Works
user/ → You define what to research
↓
agent/ → AI maintains knowledge & memory
↓
website/src/ → AI updates the Astro source files
↓
docs/ → Astro builds your website (served by GitHub Pages)
A GitHub Agentic Workflow runs on a 12-hour cron schedule. Each run, the agent:
- Reads your instructions from
user/ - Researches topics using web search (Tavily)
- Updates its knowledge base in
agent/ - Updates the Astro source files in
website/src/ - Builds the site (
cd website && npx astro build) intodocs/ - Creates a pull request with the changes (auto-merged via
auto-merge.yml)
The website is served via GitHub Pages directly from the docs/ folder.
Examples
Sites powered by AgentPages:
| Site | Topic |
|---|---|
| idorozin.me/agentpages-site | AgentPages project documentation — maintained by an AgentPages agent |
| idorozin.me/iran-monitor | Live conflict tracker — US (Epic Fury) & Israel (Roaring Lion) vs Iran (True Promise IV), ongoing since Feb 28, 2026 |
Quick Start
1. Fork this repository
Click the Fork button to create your own copy.
2. Install the gh-aw CLI
gh extension install github/gh-aw
3. Set up secrets
# Required: Tavily API key for web search gh aw secrets set TAVILY_API_KEY --value "tvly-..." # Required: API key for your chosen AI model (default is Claude) gh aw secrets set ANTHROPIC_API_KEY --value "sk-ant-..."
Get your keys:
- Anthropic: https://console.anthropic.com/
- Tavily: https://tavily.com/ (free tier: 1,000 searches/month)
4. Choose your AI model
Open .github/workflows/research.md and edit the engine block at the top of the file:
engine: id: claude # claude | codex | gemini model: claude-sonnet-4-6
All available models and their required secrets: https://github.github.com/gh-aw/reference/engines/
After changing, recompile: gh aw compile .github/workflows/research.md
5. Define your research topics
Edit user/profile.md with your interests:
## What to Research I'm interested in the latest developments in quantum computing, particularly error correction, hardware advances, and real-world applications. ## Focus Areas - Quantum error correction breakthroughs - Superconducting vs trapped-ion approaches - Enterprise quantum computing use cases
Optionally customize user/website.md for site title and style preferences.
6. Configure GitHub Pages
Edit website/astro.config.mjs and set your repo details:
site: 'https://YOUR_USERNAME.github.io', base: '/YOUR_REPO_NAME',
7. Compile and push
gh aw compile .github/workflows/research.md git add . git commit -m "Configure my AgentPages instance" git push
8. Enable GitHub Pages
Go to your repo Settings → Pages → Source: Deploy from a branch → Branch: main, Folder: /docs → Save.
9. Allow Actions to create pull requests
Go to your repo Settings → Actions → General → scroll to "Workflow permissions" → enable "Allow GitHub Actions to create and approve pull requests" → Save.
This is required for the agent to open PRs (which are then auto-merged by auto-merge.yml).
10. Run the agent
Wait for the 12-hour cron, or trigger manually:
The agent will create a PR with its first batch of research and website updates. Agent PRs are automatically merged via the auto-merge.yml workflow.
Project Structure
AgentPages/
├── user/ # YOUR files (the only folder you edit)
│ ├── profile.md # Research focus, personality, goals
│ ├── website.md # Site title, style preferences
│ ├── feedback.md # Give the agent feedback on its work
│ └── requests/ # Specific research requests
│ └── _template.md # Template for new requests
├── agent/ # AGENT's files (managed automatically)
│ ├── knowledge/ # Research findings (one file per topic)
│ └── memory/ # Action log, plan, state
├── website/ # Astro source (agent edits these)
│ ├── src/
│ │ ├── layouts/Base.astro # Shared page layout & CSS
│ │ ├── pages/ # Homepage and topic pages
│ │ └── data/site.json # Site title, tagline, topic list
│ └── public/ # Static assets (favicon, etc.)
├── docs/ # Built website (auto-generated, served by GitHub Pages)
└── .github/workflows/
├── research.md # The agentic workflow definition
└── auto-merge.yml # Auto-merges agent PRs
Separation of concerns: You only touch user/. The agent writes to agent/, website/src/, and docs/.
Giving Feedback
After the agent publishes research, you can guide it:
- Edit
user/feedback.md— tell the agent what you liked, what to change, or what to focus on - Create requests in
user/requests/— ask the agent to research specific topics - Edit
user/profile.md— shift your research focus areas
The agent reads all user files at the start of each run and adjusts its behavior accordingly.
License
MIT