openInvest — Research-Grade Investment Decision Engine¶

63⭐ open-source investment decision engine for AI agents. Instead of a single LLM making buy/sell calls, openInvest runs an isolated multi-agent committee where independent agents debate each position. Verdicts are auditable with lookahead-protected backtests. Negative results are published alongside positive ones — rare integrity in quant finance.

What It Does for Operators¶

Multi-agent committee — Each investment thesis gets debated by independent agents with different perspectives (fundamental, technical, macro, risk). No single-agent bias.
Auditable verdicts — Every decision includes the full debate transcript, evidence chain, and confidence scores. Not a black box.
Lookahead-protected backtests — Backtest engine prevents the model from "seeing" future prices when simulating historical decisions. Honest performance metrics.
Published negative results — The system publishes losing trades with the same detail as winners. No survivorship bias.
Python-native — Runs locally or on any Python environment. No cloud dependency.

Installation¶

pip install openinvest
# or
git clone https://github.com/longsizhuo/openInvest.git
cd openInvest
pip install -e .

Claude Desktop / Hermes Config¶

{
  "mcpServers": {
    "openinvest": {
      "command": "python",
      "args": ["-m", "openinvest.server"]
    }
  }
}

Key Tools¶

Tool	Description
`convene_committee`	Spawn multi-agent committee to debate a ticker/investment thesis
`get_verdict`	Retrieve the committee's final recommendation with confidence scores
`run_backtest`	Backtest a strategy with lookahead protection
`get_debate_transcript`	Full agent debate log for audit and review
`list_published_results`	All verdicts — including negative results
`analyze_position`	Deep-dive on a single position across all agent perspectives

Operator Use Cases¶

Portfolio Managers — Run investment theses through the committee before execution. Get multi-perspective analysis in minutes instead of days.
Analysts — Use the backtest engine to validate strategies with lookahead protection. Publish auditable research.
Family Offices — Automate initial screening of investment opportunities. Flag the ones worth human review.
Fintech Builders — Embed the committee engine into investment platforms. Give users AI-powered second opinions on their portfolio decisions.

Architecture¶

openInvest/
├── agents/
│   ├── fundamental.py    # Financial statement analysis
│   ├── technical.py      # Chart patterns, indicators
│   ├── macro.py          # Economic context, rates
│   └── risk.py           # VaR, correlation, tail risk
├── committee/
│   ├── convene.py        # Spawn & orchestrate agents
│   ├── debate.py         # Structured deliberation protocol
│   └── verdict.py        # Confidence-weighted aggregation
├── backtest/
│   └── engine.py         # Lookahead-protected simulation
└── server.py             # MCP server entry point

CorpusIQ Angle¶

openInvest's multi-agent committee architecture is philosophically aligned with CorpusIQ's multi-agent approach to business intelligence. Financial operators using CorpusIQ for business data (QuickBooks, Stripe, etc.) could pair it with openInvest for investment decisions — operational data feeding the investment committee. Potential deep integration opportunity.

Limitations¶

Bring your own data sources — openInvest doesn't provide market data feeds. You'll need Alpha Vantage, Yahoo Finance, or similar.
Python-only. No TypeScript/Node version.
Multi-agent committees are token-intensive. Budget ~200K tokens per full analysis.
Active development — API may change.