A daily-curated directory of AI-agent repositories — trading bots, design agents, browser automation, video generation, agent infrastructure, and more — filtered for what's actually maintained and useful, not hype.
🌐 Live site: agent-forge-nine-iota.vercel.app 📡 Daily finds on Telegram: @AgentForgeSignal
Lists like "10 GitHub repos that print money while you sleep" circulate constantly on social media. The repos they link to are often real, but the framing rarely is — README marketing language, unverified performance claims, and abandoned projects get the same spotlight as genuinely solid tools.
Agent Forge runs every candidate repo through an AI evaluator that checks for a concrete mechanism, real maintenance activity, and risk signals (closed backends behind an "open source" label, misleading "free" claims, restrictive licenses, abandoned CI, implausible timelines), then sorts the result into categories you can actually browse:
- 💰 Trading & DeFi
- 📣 Marketing & Ads
- 🌐 Browser Automation
- 🎬 Content & Video Gen
- 🎨 Design & UI/UX
- 🧠 Agent Infrastructure
- 🛠️ Dev Tools
- 📊 Data & Research
- ✅ Productivity
- 💼 Finance & Ops
Repos that don't hold up are not deleted — they're shown in a separate "hype & broken" section on the site with the specific reason, so the filtering process itself stays auditable. Nothing gets buried; you can see exactly why something didn't make the cut.
scripts/collect.py → searches GitHub (topics + keywords) and known
awesome-lists for candidate repos
scripts/evaluate.py → fetches each README, asks an LLM (via OpenRouter,
free tier with a fallback chain) to judge
verdict / category / risk flags / bilingual summary
scripts/daily_report.py → writes reports/YYYY-MM-DD.md, posts new "useful"
finds to the Telegram channel
.github/workflows/daily.yml → runs the above on a daily cron, commits
data/repos.json, and syncs it into site/data/
for the live Vercel deployment
All evaluated data lives in data/repos.json — see
docs/SCHEMA.md for the full record shape and the
verdict definitions (useful / hype / broken / duplicate).
The site itself (site/) is plain HTML/CSS/JS with no build step and a
small Three.js ambient background — it fetches site/data/repos.json and
site/data/i18n.json directly at runtime, in English and Russian.
# regenerate site data manually (requires GITHUB_TOKEN + OPENAI_API_KEY,
# where OPENAI_API_KEY is actually an OpenRouter key — see scripts/evaluate.py)
export GITHUB_TOKEN=...
export OPENAI_API_KEY=...
python scripts/collect.py
python scripts/evaluate.py
python scripts/daily_report.py
# serve the site locally
python -m http.server 8080
# open http://localhost:8080/site/index.htmlFound a repo that should be in here, or one that's flagged wrong? Open an issue with the repo link and why — the evaluator re-checks flagged entries on a future run. Pull requests for the site, the evaluator prompt, or new discovery sources (more awesome-lists, more search topics) are welcome.
MIT — see LICENSE.