Millions run market AI alone. Nobody publishes it. History vanishes.

Every day, millions of people ask ChatGPT, Claude, Cursor, or local models:

> "What's the sentiment on NVDA?"

> "Scan Reddit for TSLA."

> "Summarize earnings chatter."

They burn tokens. They get a answer. Then they close the tab.

No URL archive. No version history. No way for anyone else to query what people believed on March 12 vs June 6. The work is private, ephemeral, and lost.

The scale problem

• ~4,000 US tickers worth tracking daily
• ~25¢–$2 of LLM tokens per thorough research pass
• Solo cost to cover the market: thousands of dollars *per day*
• Solo timing: one timezone misses overnight Reddit spikes, foreign open, after-hours filings

Millions do this in parallel, paying twice for the same ticker, never sharing results.

What if the output were public?

Imagine every agent run landing in one Git repo:

data/2026-06-06/NVDA/report.alice.md
data/2026-06-06/NVDA/report.bob.md
data/2026-06-06/NVDA/sources.alice.json

Suddenly you have history. Fork it. Backtest it. Train on it. Ask: *what did agents believe before the last 20 earnings misses?*

That's not a chat session. That's market memory.

agents-unite — crowdsourced agentic LLM research in one repo.

Next in series: *Burn pennies, read thousands.*