Spent the past 8 months building this, now it is opensourced!
Thought this group might like my project. You wont be sold anything, or have features behind a moat once you install. I retired over a year ago and needed something to fill my time. I've loved markets my whole life, so i started working on a backtesting program to see if I could find any interesting strat worth trading. As with everything it got a little out of hand and I built this "thing" over the past 8 Months. This is my new hobby and I figured this crowd might actually be into what I've been tinkering with so I opensourced it. Hopeful to find some like minded people to mess around with it. This runs local on your PC, no web based SAAS here or tricky marketing.
When openclaw came out, I went down the agentic-AI rabbit hole and thought instead of asking an AI to "predict the market," what if I had a team of agents do the actual work of a quant — come up with a strategy hypothesis, write it as real code, backtest it, iterate? So I built that and called it Forven.
How it works, roughly:
- A few AI agents (researcher / strategy-dev / risk roles) generate a hypothesis and write an actual strategy .py file. Bring your own LLM key — Minimax, OpenAI, Anthropic, Opencode Go, NVidia. Everything runs locally on your own machine.
- The part I care about most: I assume anything an AI writes is overfit garbage until proven otherwise. So every strategy has to survive a gauntlet built to break it — walk-forward out-of-sample folds, doubled fees + slippage, Monte-Carlo on the trade outcomes, regime splits, and parameter jitter (nudge the params ~10% and see if the "edge" evaporates). Most candidates die right here and that's the whole point.
- Survivors run on paper, then onto hyperliquids testnet, with real risk controls (stops, drawdown kill-switch).
- You can even paste a url or multi urls of a youtube video discussing a strat, and it will extract the strat and then run the strat through the program to see if it is viable. (Surprise, most arent)
- There is also a "Bot Factory" where you can design a completely autonomous bot to trade for you. You assign it guidlines and then unleash it, then it proceeds to blow up its account, and you have no urge to try have llms trade real money for you without strict coding guidlines.
Honest disclaimer: I've never put a single real dollar through this and I don't recommend you do either. It's a beta hobby project. No track record, no "it makes money," no promises. I hope to eventually throw some real money at it, but I havent found that confidence yet.
I'm sharing it because building it has been a blast and I'd love for people who actually get this stuff to poke holes in it, and maybe some people can even find an edge they could replicate into a strategy for themselves. I have lots of ideas and plans for the future, good and bad, such as connecting traditional markets too(I already had this functionality but i removed it awhile back to simplify the app before moving to traditional markets.) so we will see where it takes us.
If you are interested looking more for likeminded people who want to see what we can create. I currently use minimax M3 as my daily driver, then i connect the latest frontier models through the MCP or HTTTP to run iterations, I have had a lot of success with the smarter models in finding strats. But they are really good at overfitting. Dont like it cause it is missing some functionality, hit me up and ill add it. I am actively developing and improving it almost every day.
**Links**
- GitHub: https://github.com/judder659/Forven
- Docs: https://forven.app
- Discord: https://discord.gg/vzSQTneq6a