Building a time machine for market intelligence

rahiakil / agents-unite

For the agentic unity

   █████╗ ██████╗ ███████╗███╗   ██╗████████╗███████╗
  ██╔══██╗██╔════╝ ██╔════╝████╗  ██║╚══██╔══╝██╔════╝
  ███████║██║  ███╗█████╗  ██╔██╗ ██║   ██║   ███████╗
  ██╔══██║██║   ██║██╔══╝  ██║╚██╗██║   ██║   ╚════██║
  ██║  ██║╚██████╔╝███████╗██║ ╚████║   ██║   ███████║
  ╚═╝  ╚═╝ ╚═════╝ ╚══════╝╚═╝  ╚═══╝   ╚═╝   ╚══════╝
  ██╗   ██╗███╗   ██╗██╗████████╗███████╗
  ██║   ██║████╗  ██║██║╚══██╔══╝██╔════╝
  ██║   ██║██╔██╗ ██║██║   ██║   █████╗
  ██║   ██║██║╚██╗██║██║   ██║   ██╔══╝
  ╚██████╔╝██║ ╚████║██║   ██║   ███████╗
   ╚═════╝ ╚═╝  ╚═══╝╚═╝   ╚═╝   ╚══════╝
  

Building the World's Financial Memory
_{Markets Change. Memory Compounds. · The GitHub of Stock Research · Open Source Alpha}

Crowdsource agentic LLM research in one repo — spend cents on one ticker, read thousands for free.

Build · Website · The Idea · Live Pulse · Join · Docs · Roadmap

The idea

Most AI agents throw away their work when the session ends. Most traders who try to LLM-research the market burn through their token budget before they finish the ticker list — and even unlimited tokens wouldn't fix timing. Asia opens while you…

View on GitHub

Something has been bothering me lately.

Every day, millions of people ask AI questions about companies, industries, earnings, and markets.

And every day, almost all of that intelligence vanishes.

A great insight appears inside a context window, helps one person, and then disappears forever.

Tomorrow, somebody else asks the same question.

And the same reasoning gets recreated again.

It feels wasteful.

Software engineers solved this problem decades ago.

They don't throw code away.

They store it.

Version it.

Review it.

Build on top of it.

So I started wondering:

What if market intelligence worked the same way?

Imagine every stock ticker having a permanent history.

Not just prices.

Not just news.

But reasoning.

Bull cases.

Bear cases.

Earnings summaries.

Macro themes.

Industry developments.

Questions people asked.

Insights AI agents discovered.

Anyone could contribute.

Use Claude, GPT, Gemini, DeepSeek, local models, or your own agent workflows.

Spend a few cents worth of tokens.

Submit a pull request.

That's it.

No single person needs to understand everything.

Nobody burns thousands of dollars in compute.

Everyone contributes tiny pieces.

And over months and years, those pieces compound into something much larger.

A living memory of how humans and AI understood companies through time.

Almost like:

GitHub × Wikipedia × Collective Intelligence × AI.

Maybe the most valuable thing isn't predicting tomorrow.

Maybe it's preserving today's understanding so future generations of humans and agents don't have to rediscover everything from scratch.

History compounds.

Memory compounds.

Perhaps intelligence should compound too.

I'd love to know if others find this idea interesting, or if there are projects I should learn from.