Multi-agent workflows
Connect research, analysis, review and notification agents — from Google Docs in, to Google Sheets and Discord out — through Your Priorities’ built-in visual workflow engine, with full audit logs and cost tracking.
Collective intelligence × artificial intelligence
Policy Synth
An open-source framework for orchestrating teams of AI agents that research problems, evolve solutions and answer with evidence — while people stay in charge of every judgment that matters.
How it works
Humans set the question. Agents do the legwork.
- 1
Define the problem — with the community
A problem statement is shaped together with the people it affects, through engagement platforms like Your Priorities and All Our Ideas, workshops or expert deliberations.
- 2
Agents research, generate and evolve
Swarms of specialized agents run thousands of search queries, map root causes, and breed populations of candidate solutions over many generations — pairwise-ranked with Elo ratings, each with researched pros and cons.
- 3
People review, vote and decide
Every output flows back to human review — spreadsheets, voting rounds, workshops. AI accelerates the work between human judgments; it never replaces them.
Capabilities
One framework, many kinds of agents
Evolving solutions
Genetic algorithms breed and refine solution populations across generations, with pros, cons and evolutionary trees for every idea.
Deep automated research
Over a thousand generated search queries per problem — general, scientific, data and news — distilled into root causes and evidence.
Grounded chatbots
A full RAG stack with ingestion, Elo-ranked chunking and validation agents that retract hallucinations by majority vote — powering ERIC and the Rebooting Democracy bot.
Multi-model by design
Run the same analysis through GPT, Claude and Gemini side by side and compare — trust through triangulation, not faith in one model.
Human in the loop
Voting platforms, review sheets and workshops are first-class agent nodes — collective intelligence is part of the architecture, not an afterthought.
Use it two ways: standalone via the @policysynth/agents TypeScript library, or through the visual agent workflow engine built into Your Priorities — the interface you see in our case studies.
In the field
Already making policy, today
New Jersey AI Task Force
2,200+ workers ranked 96 concerns about generative AI; Policy Synth turned the top 20 into evidence-based policy solutions.
Read the story →Iceland — Gold-plating
Scanning Icelandic law against EU directives, article by article — 52 unjustified additions found in the telecoms pilot alone.
Read the story →Skills-First Hiring
Thousands of state job descriptions analyzed and rewritten to remove unnecessary degree requirements.
Read the story →Countering Election Subversion
With Democracy Fund and The GovLab — the first real-world test of AI-accelerated Smarter Crowdsourcing.
Read the story →ERIC — European Union
An AI rights assistant for European citizens, grounded strictly in official legislation.
Read the story →Unlocking Literacy — Boston
Decomposing the reading gap and evolving evidence-based solutions with Boston schools and the Museum of Science.
Read the story →Build with Policy Synth
MIT-licensed TypeScript, published on npm — the same packages we run in production.