Open source and credits
We learn from open work and say so. Where we studied a method, we wrote our own code; no code is copied unless it says so here.
Our open repositories
- playfellow-coop The accessibility co-op: voice commands, hand-over, marks, benchmarks. Open source.
- playfellow The research engine the co-op builds on.
Studied and re-implemented
- Elephant-Goldfish Model (Google) Our review habit: a fresh reader checks every piece of work against its contract. Apache-2.0.
- Harness² (Google Research) Propose, probe and compose: improving the harness around a fixed model. Apache-2.0.
- Perfetto (Google) We export our traces in a format Perfetto opens. Apache-2.0.
- Agent Development Kit (Google) Agent evaluation and tooling, used outside the real-time loop. Apache-2.0.
- meta-sam and the Meta Model API cookbook How we use the Model API: schema replies, prompt caching, retry rules, and reading SAM's text format.
Research that inspired Playfellow
Inspired by research such as Google DeepMind's SIMA 2 and NVIDIA's NitroGen. There is no affiliation or endorsement.
Fonts: Baloo Bhaijaan 2 and Rubik, under the SIL Open Font License 1.1. Built with AI coding assistance, including Meta's Muse Code.