
Founded 2026 · Open source · Apache 2.0
Forger Labs HQ
We build open-source tools that make AI-assisted research reproducible, recorded and ready to ship.
Why we build it
Most ML experiments disappear.
They live in a notebook someone ran six months ago, in a thread nobody can find, in a note that was never written down. The winning idea ships; everything that failed, and why, is forgotten.
The failures are where much of the knowledge is. Knowing why something did not work is as useful as knowing what did.
ResearchForge keeps that record. It freezes a baseline, runs each idea in isolation, keeps every result and every rejection with its reason, and ships only what holds up.
What we hold to
Evidence first
Nothing counts as a win until it beats a frozen baseline, and failed ideas stay on record with the reason they failed.
Open source
The whole loop is Apache 2.0, with no lock-in and no hidden service. Use it, fork it, build on it commercially.
AI proposes, people decide
Claude Code, Cursor or a model you choose does the heavy lifting. You approve the contract, the plans, and anything that ships or submits.
Team
Manas Joshi
Founder
Built ResearchForge after running experiments for a $50k Kaggle competition and finding no good way to track what was tried, what failed and why. The missing tool became the first product of Forger Labs HQ.
Forger Labs HQ is very early, started in 2026. If you want to build with us, get in touch.
Follow the work on GitHub.