Private inference
Exploring model execution that keeps sensitive context close to the user.
AI research lab
Private local inference, agent harnesses, and play.
PumaAI explores how useful AI can run closer to the user: private local inference, multi-agent harness flows, and a side of fun gaming experiments.
Local inference
Agent harnesses
Research Bench
We prototype local model runtimes, agent orchestration loops, and game-like testbeds that make AI systems easier to inspect, steer, and enjoy.
Inference
On-device paths
Harness
Agent flow tests
Games
Playable probes
Notes
Open questions
PumaAI today
A small AI lab testing what private, local-first systems can become.
Local-first systems
Testing practical ways to run useful AI on personal hardware.
Multi-agent harnesses
Building loops that make agent behavior observable, repeatable, and easier to debug.
Flow experiments
Prototyping handoffs, tool calls, evaluations, and human-in-the-loop controls.
Gaming experiments
Using playful prototypes as a fast way to stress-test agents and interfaces.
Research notes
Turning prototypes into concrete lessons about privacy, control, and agency.
Long-term vision
We believe personal AI should be private, steerable, and genuinely enjoyable.
Private compute
More intelligence should run locally, with cloud services used deliberately.
Agent workflows
Multi-agent systems need harnesses that expose decisions, failures, and tradeoffs.
Playable testbeds
Games can make complex AI behavior legible, repeatable, and fun to improve.
Values
Privacy, agency, curiosity, and play.
Privacy
Agency
Curiosity
Play
Backed by builders
Angel and pre-seed support from leaders in crypto and AI.
Stefan Thomas (Interledger, BitcoinJS), Chris Larsen (Ripple), Anatoly Yakovenko (Solana), Illia Polosukhin (NEAR, Attention paper), Sridhar Ramaswamy (Snowflake, Neeva), Don Ho (Quantstamp, Orange DAO), Kartik Talwar (ETHGlobal, A.Capital), Jason Warner (poolside.ai, GitHub), Oleksandr Maksymets (Meta Superintelligence Labs), Evil Rabbit (Vercel), and many others.
Join the team
We are researching private local inference, multi-agent harnesses, and game-like AI experiments. If that sounds like your kind of lab, we want to meet you.
Engineering
Research
Games
Join at puma.tech
Puma is hiring
We are focused on making local AI practical, inspectable, and fun.
- Remote-friendly team
- Private local inference
- Multi-agent harness flows