10 AI investing repos worth watching right now
Not the ten biggest. The ten where the next six months will tell you something: fastest growers, execution engines going AI-native, and the infrastructure that agents are converging on. Live numbers, updated daily.
In this article 11 sections
This list is deliberately not sorted by stars. The leaderboard and the radar already do that, every day, without our help. Instead, these are the ten repositories where we think watching the next few months of commits, releases and star velocity will tell you something about where AI investing is going. The table is live; the commentary was written on 2026-10-09.
- Stars
- 110,565
- Stars gained · 7d
- +889
- Stars gained · 30d
- +6,452
- Last push
- 8 days ago
- License
- Apache-2.0
- Stars
- 32,243
- Stars gained · 7d
- +119
- Stars gained · 30d
- +620
- Last push
- 19 days ago
- License
- Other
- Stars
- 13,032
- Stars gained · 7d
- +50
- Stars gained · 30d
- +211
- Last push
- 7 days ago
- License
- AGPL-3.0
- Stars
- 12,660
- Stars gained · 7d
- +238
- Stars gained · 30d
- +1,242
- Last push
- 2 days ago
- License
- Apache-2.0
- Stars
- 7,276
- Stars gained · 7d
- +75
- Stars gained · 30d
- +270
- Last push
- today
- License
- AGPL-3.0
- Stars
- 14,893
- Stars gained · 7d
- +57
- Stars gained · 30d
- +334
- Last push
- 5 days ago
- License
- MIT
- Stars
- 29,839
- Stars gained · 7d
- +251
- Stars gained · 30d
- +1,194
- Last push
- today
- License
- LGPL-3.0
- Stars
- 2,117
- Stars gained · 7d
- +18
- Stars gained · 30d
- +74
- Last push
- today
- License
- GPL-3.0
- Stars
- 1,016
- Stars gained · 7d
- +13
- Stars gained · 30d
- +69
- Last push
- 16 days ago
- License
- MIT
- Stars
- 1,427
- Stars gained · 7d
- +4
- Stars gained · 30d
- +16
- Last push
- 2 days ago
- License
- MIT
| Project | Stars | 7d | 30d |
|---|---|---|---|
| TradingAgents | 110,565 | +889 | +6,452 |
| TradingAgents-CN | 32,243 | +119 | +620 |
| NOFX | 13,032 | +50 | +211 |
| QuantDinger | 12,660 | +238 | +1,242 |
| OpenAlice | 7,276 | +75 | +270 |
| RD-Agent | 14,893 | +57 | +334 |
| NautilusTrader | 29,839 | +251 | +1,194 |
| Lumibot | 2,117 | +18 | +74 |
| Alpaca MCP Server | 1,016 | +13 | +69 |
| TradeMemory Protocol | 1,427 | +4 | +16 |
1. TradingAgents
The reference multi-agent framework and still the fastest-growing repository we track in absolute terms, at roughly +889 stars a week. Watch it for two reasons: it is where the "analysts, debate, trader, risk" pattern gets refined first, and its README has become unusually careful about what the backtests do and do not prove. If the project ever publishes a cost-per-decision figure or a live record, the whole category will have to follow.
2. TradingAgents-CN
A fork with 32,243 stars is not a fork, it is a second project. The Chinese-language extension of TradingAgents targets A-shares, adds domestic LLM and data providers, a web UI and paper trading, and has a hybrid licence with source-available components. It is the clearest signal that a large share of open-source AI-trading activity, and of the audience, sits outside the US market and the English-language discussion.
3. NOFX
Created in late 2025, already past twelve thousand stars, written in Go, and the most aggressive project on this list about actually trading: nine exchange integrations, a continuous trade loop and a runtime risk engine that clamps every order. It is also the project whose documented first run deposits USDC on a perpetuals exchange, with no paper mode in the README. Watch it as the test case for whether "AI terminal" products can grow up into something with a sandbox and a track record.
4. QuantDinger
The other big self-hosted "AI trading OS", and in some ways NOFX's opposite: deterministic Python strategies and risk checks run the show, and the AI sits in front of live entry orders as an optional, fail-open gate. It needs Docker, Postgres, Redis and Kafka, which tells you who it is for. Its MCP agent gateway is one of the first examples of an execution platform exposing itself to external coding agents rather than embedding its own.
5. OpenAlice
A TypeScript, local-first workspace that hosts coding agents (Claude Code, Codex and others) inside Git-backed research workspaces and makes trade execution a pull request: agents stage proposed operations, a human approves. It is young and the execution layer is marked beta, but "trading as Git" is the most interesting answer we have seen to the question of how much autonomy an agent should get.
6. RD-Agent
Microsoft's research-and-development agent is on this list because it is the one large project that applies LLM agents to the quant's actual job, discovering factors and models in Qlib, rather than to placing trades. It is Linux-only, needs Docker and a model with embeddings, and its README claims about twice the annualised return of benchmark factor libraries without stating a period. Watch for the paper's methodology to make it into the README, and for anyone replicating the result.
7. NautilusTrader
Not an AI project, and that is the point. A Rust-native, event-driven trading engine with identical backtest and live code paths and +251 new stars a week is the kind of infrastructure that agent frameworks eventually need when they stop printing decisions and start routing orders. The question is which agent project integrates it first.
8. Lumibot
Six years old, GPL-licensed, twelve broker integrations, and since 2026 repositioned around "AI agents that actually place the trade." Agents are created inside a normal strategy class, a deterministic Python gate sits between them and the broker, and decisions are written to SQLite and Parquet. The README is also an honest record of the state of the art: one recorded agent run, one verified fill of 15 SPY shares, and the sentence "a fresh model run can choose different trades and returns."
9. Alpaca MCP Server
The official broker MCP server, rewritten this year from the OpenAPI spec, paper-by-default, with server-side toolset restriction. It matters less for its star count than for what it represents: a regulated broker treating AI assistants as a first-class client. Watch which other brokers follow and whether any of them ship a remote, OAuth-based server the way Twelve Data has on the data side.
10. TradeMemory Protocol
A memory and audit-trail layer for trading agents, delivered as an MCP server, with a preview "brake" that sits between an agent and Alpaca as a policy proxy. The maintainers publish their own validation status ("n=14 trades, target n≥100") rather than a return. If the multi-agent frameworks are the brains of this ecosystem, projects like this are its conscience, and we would like to see more of them.
Honourable mentions
AI Hedge Fund for being the most-starred project that refuses to trade; FinRobot for moving the arithmetic out of the model and into deterministic operators; edgartools for making SEC filings a Python object; and Kalshi AI Trading Bot for publishing "no measured edge" on its own live account, which is more than most of the industry has done.
The numbers above come from the same daily sync that feeds the rest of the site, so this page will quietly disagree with the commentary over time. That is the intended behaviour.
Cite this research
InvestingStack (2026). 10 AI investing repos worth watching right now. InvestingStack. https://investingstack.ai/research/ai-investing-repos-worth-watching
Published: Oct 9, 2026
