Produce cited research briefs with simulated peer review
Feynman multi-agent system searches papers and the web, synthesizes findings, runs simulated peer review, verifies citations, and outputs a cited brief.
Done with a custom-built or unnamed agent
- What they did
- User runs one curl-bash install from terminal then types a topic. Researcher agent pulls evidence from papers, repos, docs, and web; Reviewer runs simulated peer review with severity-graded feedback; Writer drafts paper-style output from notes; Verifier checks every citation and drops dead links. Additional agents can replicate experiments on local/cloud GPUs or audit papers against codebases.
- How it went
- Produces a fully cited research brief with all claims source-grounded. No usage metrics or shortfalls reported in the post.
- Worth knowing
- Single curl install command; repo at github.com/Companion-Inc/feynman; thread notes the installer script is not itself open.
Try it yourself with Manus
Research [topic] and write a cited brief of about [length]. Search academic papers and the web, synthesize the main findings, then critique your own draft as a skeptical peer reviewer and fix the weak points. Verify every citation actually exists and supports its claim; you're done when I have the brief with a reference list and a note on any claims you couldn't verify.
Source: X · Apr 6, 2026
Five of these in your inbox every morning
The best things people got an AI agent to do, each with the prompt to try it.
More like this
Find connections across 100 non-fiction books
Claude Code browses a personal library with CLI tools and surfaces trails of related excerpts across books.
Scrape weekly Substack rankings and write a report
Claude in Chrome runs weekly to scrape Substack ranking data into a dataset, and Cowork analyzes it and writes a weekly report.
Summarize 20 articles into a cited document
An agent extracted key insights from 20 articles, found common themes and produced a cited summary; the write-up estimates 18 minutes versus hours by hand.
Research conference speakers and organize tabs
Comet researched 50 HumanX speakers, surfaced themes and podcast guests, built a camera spec table, and cleaned up tabs.
Let Cursor maintain an Obsidian notes vault
Cursor reviews vault structure, suggests connections, answers questions about notes and writes a daily done note documenting other agent sessions.
Research content from Slack and draft Notion guide ideas
A team DM'd Lindy in Slack to summarize a year of guides and research a model, and it created a Notion guide idea after asking for approval.