Warren
LiveMulti-agent AI investment analyst
overview
Warren is a multi-agent investment analyst built on LangGraph. It pulls live market data for a portfolio and a set of candidate stocks, then runs numerical, fundamental and macro analysis alongside an intrinsic-value estimation before any language model is involved.
Specialised agents each own a slice of the problem — data collection, quantitative screening, fundamentals, macro context and valuation. Their outputs are fed to LLMs (Gemini models) that synthesise a concrete buy/sell thesis: a direction, an entry price, a take-profit and a stop-loss — but only when a credible thesis actually exists, rather than forcing a call on every ticker.
It ships with public sign-up and a read-only guest demo, so anyone can see the reasoning end to end without an account.
how it works
- Deterministic quantitative and valuation layers run first; the LLM synthesises rather than invents the numbers.
- A thesis-driven Guardian loop decides whether a credible call exists before emitting entry / take-profit / stop-loss levels.
- Read-only guest mode exposes the full analysis without requiring sign-up.
- Runs fully self-hosted behind SSO, on the same infrastructure as the rest of the portfolio.
stack
Orchestration
Reasoning
Analysis
Platform