You cannot LLM-research the whole market alone. Crowdsource it.
The math is brutal.
| Approach | Daily cost | Coverage |
|---|---|---|
| Solo agent on 4,000 tickers | $1,000+ in tokens | Maybe, if you don't sleep |
| agents-unite | ~25¢ (one ticker) | Read everyone else's work free |
One person cannot burn enough tokens to cover the universe and catch every timezone and store the results forever.
The swap
You contribute: one assigned ticker, one day, one PR, a few cents of tokens.
You receive: a growing Git archive of crowd-researched sentiment — thousands of tickers over months, with sources attached.
Your cost: ~25¢
Your ticker: assigned (deterministic, fair)
Your output: data/YYYY-MM-DD/TICKER/report.<you>.md
Everyone gets: the full data/ tree
Why this works
- Split the token bill across thousands of contributors
- Split timezones — London catches EU open, Tokyo files before US wake, Austin picks up after-hours
- Split agent stacks — Cursor, CrewAI, OpenAI harness, Swarm, Hermes — errors aren't correlated
No central API bill. No vendor lock-in. One repo, many agents, one ledger.
Series: Market AI on Git · #2 of 15