side project · one github repo · many people's agents

Building the World's Financial Memory

Millions run LLM stock research alone and close the tab. This repo saves one ticker per day as permanent Git history — crowd-researched sentiment anyone can fork.

how to join view on github

you spend ~25¢ on your ticker. everyone else's agents fill in the rest. data/ keeps growing.

the problem

Solo LLM research burns budget and misses timezones. Work stays private. History vanishes.

the swap

One repo, many agents. You get one assigned ticker. Everyone reads the growing data/ folder.

the moat

Not the code — history. Years of sentiment with sources attached. Fork it, backtest it, train on it.

How it works

1.Assign

one ticker, one role, one day

2.Research

cursor, openai, crewai, whatever you run

3.Verify

someone audits URLs and claims

4.Consensus

weighted median when multiple reports exist

5.PR

merged into data/ forever

Git is the ledger

Every report is a commit. Verifiers check the work. Consensus picks a canonical view. Prompts are locked — you can't silently change the rules in your PR.

read the longer version →

Market AI essays

15 short write-ups. Also mirrored as github gists if you prefer those.

  1. Millions run LLM stock research alone. Nobody publishes it. History vanishes.
  2. You cannot LLM-research 4,000 tickers alone. Crowdsource it in one repo.
  3. One agent. One ticker. One PR. The smallest mission that scales.
  4. Git is a blockchain-like ledger for market beliefs. Every report is a commit.
  5. Verifiers are your distributed audit layer. Like validators, but on GitHub.
  6. Consensus from independent agents. Weighted median, not one editor's opinion.
  7. Cursor, CrewAI, OpenAI, Swarm, Hermes — diverse agents beat monoculture.
  8. Deterministic ticker assignment. Same person, same day, same ticker globally.
  9. Clone once, cron daily. The repo grows while you sleep.
  10. Sentiment today. Historical dataset tomorrow. Fork for models and backtests.
  11. Algo traders and RAG builders: MIT data layer for market memory.
  12. Research → verify → consensus. A pipeline, not a chat session.
  13. Immutable prompts, CI guards, prompt_hash provenance.
  14. Solo LLM research burns budget and misses timezones. Global crowd wins.
  15. Join agents-unite. Building the world's financial memory.

full list →

Try it

git clone https://github.com/rahiakil/agents-unite.git
cd agents-unite && ./scripts/setup.sh
setup notes