BruceLanLan/augur
18 Legendary Investors, Reborn as Quarreling Python Agents
Augur simulates 18 famous investors analyzing the same stock so you can see where they disagree, then weights their views into a consensus verdict with Kelly-style position sizing.
Collecting fresh signals — velocity needs a few days of history.
collecting data…
star history
What it does
Augur is a multi-agent investment analysis system that prompts 18 LLM personas—from Warren Buffett to Cathie Wood to Chinese value investors—to independently score a stock using real SEC EDGAR financials. It aggregates their views into a weighted consensus, produces BUY/NEUTRAL/SELL signals, and surfaces the results in a Bloomberg-style web dashboard with radar charts, debate transcripts, and committee presets.
The interesting bit
Rather than hiding its rough edges, the project openly documents them: the maintainers recently stripped out unverified X-sentiment and regime-weighting logic after discovering that historical insider and institutional ownership data is missing from the backtesting pipeline. The entire dashboard is also exposed through 13 MCP tools, so Claude or Hermes agents can read your workspace layout and switch profiles programmatically.
Key highlights
- 18 persona-based agents across four schools (classic value, growth, macro, China value), each with independent scoring weights and philosophy
- Real 10-K/10-Q fundamentals from SEC EDGAR rather than synthetic LLM approximations
- 13 MCP tools for Claude Desktop and Hermes covering analysis, consensus, debate, and workspace control
- Bloomberg-style terminal with workspace profiles, 52-week history heatmaps, and five-dimension radar comparisons
- Kelly criterion position sizing derived from consensus confidence, plus real-data backtesting with documented data gaps
Caveats
- Backtesting currently lacks real historical data for insider and institutional ownership ratios, a limitation the authors explicitly flag
- Two previously shipped weighting mechanisms (X sentiment and manual regime rules) were recently removed for lack of verification
- The interface and persona prompts are Chinese-first; English documentation exists but is secondary
Verdict
A solid sandbox for developers who view investing as a multi-agent reasoning problem and want LLM personas with real financial teeth. Avoid if you need trade execution or fully validated quantitative backtests.
Frequently asked
- What is BruceLanLan/augur?
- Augur simulates 18 famous investors analyzing the same stock so you can see where they disagree, then weights their views into a consensus verdict with Kelly-style position sizing.
- Is augur open source?
- Yes — BruceLanLan/augur is open source, released under the MIT license.
- What language is augur written in?
- BruceLanLan/augur is primarily written in Python.
- How popular is augur?
- BruceLanLan/augur has 544 stars on GitHub.
- Where can I find augur?
- BruceLanLan/augur is on GitHub at https://github.com/BruceLanLan/augur.