// the find
AI4Finance-Foundation/FinRobot
FinRobot: An Open-Source AI Agent Platform for Financial Applications using Large Language Models
FinRobot is an agent framework for automating equity research — pulling market data and filings, running DCF/DDM/LBO valuations, and producing research reports — built by the AI4Finance Foundation (also behind FinGPT). It's aimed at developers who want to automate parts of investment research, not day-to-day traders looking for signals.
Numbers come from pure-Python compute operators (DCF, DDM, LBO, WACC, comps), not the LLM, and every output is provenance-tagged — the one design decision that actually matters for a finance tool and they got it right. The debate-agent step (bull vs. bear vs. judge) before producing a verdict is a real mechanism, not agent theater. The data layer has 7 providers with failover (FMP, Finnhub, yfinance, SEC EDGAR, etc.) instead of a single API dependency. All three generations of the system — AutoGen, OpenAI Agents SDK, PydanticAI — are still in the repo and runnable, so you can read the actual architectural evolution instead of just the current state.
The repo is really three different codebases (different agent frameworks, Python + Rust + React) stapled together, and figuring out which one to clone takes effort — the one on PyPI (`pip install finrobot`) is explicitly the one they tell you not to use for real work. The desktop app is macOS Apple Silicon only and isn't notarized, so first launch requires a manual `xattr -cr` to get past Gatekeeper; everyone else is routed to an unauthenticated local web UI instead. There are no evaluation numbers anywhere in the README — no backtests, no accuracy against analyst consensus, nothing to back up 'decision-grade' other than the claim itself. The README also doubles as a funnel into the paid FinRobot Pro hosted product, so it's not purely an open-source pitch.