Agent Radar Scheduled Markdown research
Get sourced briefs on AI agents, kept in Git.
A model drafts the Markdown reports from public sources on a GitHub Actions schedule.
python scripts/agent_radar.py init -
radar.md -
agent-watchlist.md -
user-field-notes.md -
playbook.md -
storage-angle.md
- GitHub Actions Python, GitHub Actions
- Online connects on its own, see Privacy
- MIT open-source licence
- v0.26.3 latest release
What the Markdown files hold
-
Current thesis
In
radar.md: the current view, how it changed, and open questions. -
Watchlist
Mainstream and emerging agents, each source graded from Strong to None.
-
Field notes
Real user workflows, complaints, tricks and failures, plus a playbook of reusable patterns.
-
Storage implications
What agents mean for workspaces, sandboxes, logs, and replay.
Get started
Run the commands from a clone of the repository. They need Python 3.10 or later and only its standard library.
-
Create the Markdown files
python scripts/agent_radar.py init -
Create or add to today’s daily note
python scripts/agent_radar.py daily
FAQ
Does it need paid search?
No. It collects from free public sources and calls no paid search service; you choose the AI model service.
- The setup pinned in the workflow uses GPT-5 Mini to screen sources and draft reports.
- That costs about $2.60 a month, against a $4 budget.
- Its fallback models are free ones.
Where does it run?
Daily, weekly, and monthly jobs run on GitHub Actions and produce bilingual reports. If a Chinese half comes out thin, the runner regenerates it, or labels the report when it cannot. Local runs are available for debugging.
What if a check fails?
Most checks repair or label instead of refusing: a missing section gets an explicit gap line. Failures stay visible in the run telemetry, in warnings from validate, and in GitHub issues; Git records what was published.