Compare
Chart Library vs Koyfin — a research terminal for a human versus a primitive for an agent.
Koyfin is an excellent Bloomberg-lite research terminal: charts, financials, estimates, dashboards, watchlists and screens, all in a polished UI for a human analyst to explore. If you sit and research, it’s a great seat.
Chart Library isn’t a UI you sit in. It’s an endpoint your agent calls: hand it any (symbol, date, timeframe) and it returns the cohort of historical analogs, what they did next, and a calibrated outcome distribution with an audit receipt. A research terminal helps a person form a view; Chart Library hands a program the calibrated base rates to reason on.
Human research UI vs agent primitive
- Koyfin = exploration for humans. Dashboards and charts you read and interpret yourself. The synthesis happens in your head.
- Chart Library = a callable answer. One MCP/REST call returns a structured, calibrated outcome distribution + drivers + receipt — the synthesis is done, machine-readable, and citable.
- The calibration layer is the difference. A terminal shows you data; Chart Library ships the trust receipt: nominal 80% band held 80.8% across 300K+ audited cases, PIT-flat. Never a forecast.
- Use both. Research a name in Koyfin; have your agent pull the calibrated historical base rates for the specific setup from Chart Library.
Frequently asked questions
- Is Chart Library a Koyfin alternative?
- Only if your 'user' is an agent rather than a person. Koyfin is a research terminal a human drives; Chart Library is an API/MCP an agent calls for calibrated historical-analog outcomes. Different audiences and form factors.
- Does Chart Library have dashboards, screens, and fundamentals like Koyfin?
- No. It's a focused retrieval-and-calibration engine for analogs and outcome distributions, built to be called by code. For research dashboards, use Koyfin; for the calibrated base rates on a setup, call Chart Library.
- Does it tell me which stocks to buy?
- Never. It returns what historical analogs of a setup did next as a calibrated distribution with a coverage receipt — descriptive, not directional.
Try it
Run a cohort_analyze call.
Free Sandbox tier — 1,000 calls/day, no authentication. MCP install for Claude or Cursor takes 30 seconds.