Strands Labs is the experimental arm of
Strands Agents: open source projects that take the core SDK into new problem
spaces. Where the SDK gives you the agent loop, tool use, model providers, and
multi-agent patterns, Labs applies that foundation to areas like physical
robotics, world models, agentic benchmarking, harness optimization, and
real-time audio.
Labs projects are published to package repositories and work alongside the SDK
today. They move faster and cover more surface area than the core SDK, so expect
more frequent changes and newer integrations. Some projects graduate into the
core SDK or become standalone products; others stay experimental. Each project
lives in its own repository under the
strands-labs organization.
RobotsControl, simulate, and train physical robots with natural language. One Robot() call returns a MuJoCo simulation or real hardware, with pluggable vision-language-action policies and a peer-to-peer mesh.
Strands for CosmosBring NVIDIA Cosmos to Strands Agents: physics-aware reasoning over video, plus generation of video, audio, and robot actions on local compute.
Benchmark HarnessesStrands-based agents and harnesses for agentic benchmarks, including Simple Strands Agent, a lean autonomous-coding harness with strong results on SWE-Bench and Terminal Bench 2.
Harness OptimizerOptimize an agent's harness through tunable Formulas, then improve those Formulas from collected rollout trajectories with a PyTorch-style training loop.
AI FunctionsPython functions evaluated by AI agents. Enforce correctness with runtime post-conditions instead of prompt engineering alone, and compose functions into multi-agent workflows.
PyWebRTC AudioPython bindings for WebRTC audio processing: echo cancellation, noise suppression, gain control, and voice activity detection, with a working Strands BidiAgent integration.