Ideate & research
Continuous exploration of emerging technologies.
THE LAB · R&D
Where our tools come from.
We prototype new tech before it touches a client project. What survives gets released as open source or folded into our delivery stack.
Recent research
SkillGym — 4M Labs Research
Public Agent Skills Are Becoming the Next RL Dataset
A pipeline that compiles public agent skills into verified training environments for RL fine-tuning, proving that agent skills transfer measurable reasoning gains to small language models.
1.5BParameters
13Skills
80Tasks
99SFT Examples
67DPO Pairs
What comes out of the Lab
- 01SkillGym — verified training environments for agent skills
- 02Squish — persistent memory for coding agents
- 03AI agent frameworks
- 04Reusable automation primitives
- 05Research papers
- →More research shipping soon.
How we work
Prototype fast
Functional prototypes to validate feasibility.
Release open source
Reusable components back to the community.
2-3Weekly prototypes
15+Open-source projects
6,000+Pip installs / month
8Working papers
Collaborate with the Lab
Have a research problem, or want to contribute to what we release?
Get in touchFAQ
What types of projects does the Lab take on?
The Lab focuses on AI and automation work that explores new technology and produces reusable components. We build agent frameworks, open-source tools, and automations that speed up Studio and Delivery Engine builds.
Do you take equity?
Yes. For equity-based engagements we negotiate a stake. The Lab takes shared risk on selected projects in exchange for a piece of the long-term upside.
How do you select projects?
Three criteria: real demand, technical feasibility on our stack, and scale potential beyond the initial prototype.
Can I bring my own idea?
Yes. Send us your concept and we review it against our selection criteria. Ideas with potential can become joint incubations.