Add historical pattern similarity as a data source for the technical analyst agent

[grahammccain]

Hey — love what you've built here. The multi-agent architecture with specialized analyst roles is a really clean design.

I built Chart Library, an API that returns historically similar chart patterns + their forward returns. Think of it as "what happened the last 10 times a stock looked like this." It could be a natural data source for the technical analyst agent — giving it historically-grounded pattern context instead of relying only on indicator calculations.

Quick integration example:

import requests
# One call returns 10 similar patterns + 1/3/5/10-day forward returns + regime context
resp = requests.get("https://chartlibrary.io/api/v1/intelligence/AAPL")
pattern_data = resp.json()
# e.g. "7 of 10 similar patterns were higher 5 days later, median +2.1%"

There's also an MCP server (pip install chartlibrary-mcp) if you want the LLM agents to call it directly as a tool.

Free tier is 200 calls/day, no API key needed. 24M pre-computed embeddings across 15K stocks and 10 years of data.

Happy to help with integration if there's interest. No pressure either way — just thought it was a natural fit for the technical analyst role.

[JohnnyTarrr]

Complementary to chart pattern matching — we provide the fundamental context that explains why patterns resolve the way they do. Our API returns SEC EDGAR filings, CFTC futures positioning, social sentiment (X + Reddit), and Wikipedia attention spikes alongside price data.

For example: "NVDA attention spiked 3x on Wikipedia last time before a 12% move" or "hedge funds went net short S&P ahead of the last correction." Pattern recognition tells you what happened before — we add why and what's different this time.

Integration: TradingAgents PR #439 or TradingAgents-Pro.

[grahammccain]

This is a great complement — we've actually been thinking about exactly this pairing. Pattern matching tells you "this setup appeared 10 times and 8 went up," but it can't tell you why it resolved that way. Fundamental context (positioning, sentiment, attention) is the missing layer.

The interesting question is whether combining both improves the signal. If 8/10 similar patterns went up, but this time hedge funds are net short and sentiment is inverted from the historical matches — that's probably worth knowing before sizing a position.

I'll check out TradingAgents PR #439. Would be cool to see a multi-agent setup where one agent pulls pattern context from Chart Library and another pulls fundamental context from your API, then a third synthesizes.