Toolbox/Applied agents

Robotics & embodied

agents that plan, code and debug for robots and simulators

16 repositories, most starred first
State of the category

nasa-jpl/rosa nasa-jpl/rosa is the de facto leader for teams who want tight natural-language integration with ROS systems—pick it if you need easy LLM-driven inspection, diagnosis, and control of ROS1/ROS2 robots. If you need different trade-offs (MCP-standard bridges, audited runtime safety, edge/offline operation, or turnkey GUIs and benchmarks) other projects target those gaps.

MCP & bridging
Many teams need a standardized, no‑source-code way to let LLMs fully observe and control robots across different fleets and stacks, so multiple projects implement MCP-style bridges and lightweight servers.
Runtime safety & audit
Production robots need enforced safety layers, auditable execution and fail‑closed guarantees so operations can be trusted in the real world.
Benchmarks & training
Teams building code-as-policy agents require benchmarked environments, dataset recording and fine-tuning tooling to evaluate and improve long‑horizon behavior.
ROS-native & edge-first
Some projects prioritize running directly on robots or entirely offline with ROS2-native stacks and local multimodal models to meet latency, privacy, and connectivity constraints.
Developer ergonomics & turn‑key stacks
Practical adoption is driven by production prompts, developer-facing SKILLs, GUIs and out‑of‑the‑box agent packs that make experiments and deployment faster for engineering teams.
Find your fit

Which one matches your setup?

Answer any of the questions — the shortlist updates as you go. Recommendations come from the capability passports below, nothing else.

Where should reviews happen?
Can code leave your infrastructure?
What matters most?
Model access?
Pick at least one answer to get a shortlist.
Side by side

Comparison matrix

Axes are extracted from each project's docs by our review pipeline; the maturity score is computed from stars, growth and commit activity — not an opinion. Click a column to sort.

Runs inModelsContextCost to run
1.4k
CLIBYOKOpenAIAnthropicGeminiFull robot contextMinimal guidanceSelf-hostableDepends on external model providers; uses your model client / API key.
1.0k
CLIFixed providerWorkspace & protocolsMulti-layer validationSelf-hostableFree, self-hosted
555
CLIBYOKDiff onlyNoneSelf-hostableOpen-source, self-hosted; model API costs may apply
322 +6/7d
Coding-agent pluginCLIWeb appIDEBYOKAnthropicFixed providerPrompt-only contextNo noise controlsSelf-hostableFree skill files; model/agent API costs may apply
661
CLIWeb appCIBYOKGeminiFixed providerDiff onlyNone mentionedSelf-hostableYour API key or local GPU
305
CLIFixed providerRepo + related filesIsolation & recoverySelf-hostableSelf-hosted (hardware/runtime costs)
170
CLICIWeb appOpenAIAnthropicGeminiDiff onlyBasic filteringSelf-hostableFree, self-hostable; optional model API costs
124
CLIWeb appGeminiSingle-script contextNo explicit controlsCloud API (your key)Requires cloud LLM access (your API key).
105
CLIIDEBYOKTool-onlySim-safe defaultsRuns fully localFree; local GPU required for some backends/training
65
CLIWeb appOllama / localFixed providerGraph-level contextNo explicit controlsFully local / offlineFree — runs locally on your hardware (no cloud required)
1.6k
CLIBYOKSystem-level contextNone mentionedSelf-hostableYour LLM/API key (your usage costs)
294
CLIWeb appBYOKOpenAIAnthropicGeminiOllama / localFixed providerDiff/file onlyPydantic validationRuns fully localYour API key for cloud models; optional hosted endpoints.
Ranked by maturity — 4 more in the full catalog below.
At a glance

Capability profiles

Six axes, 0–5 each. The shape tells you the strategy: a wide hexagon is a generalist, a spike is a specialist. Showing the 8 most established — the rest are in the full catalog.

ContextNoiseCustomPrivacyModelsMaturity

Bridges ROS and the MCP standard to let LLMs fully observe and control robots without modifying robot source code.

ContextNoiseCustomPrivacyModelsMaturity

Decouples cognition from physical hardware with a session-centered runtime and adapter/bridge architecture that lets the same sessions run unchanged across simulation and real robots while enforcing multi-layer safety and auditable execution.

ContextNoiseCustomPrivacyModelsMaturity

A ROS2-native, vendor-agnostic embodied agent framework focused on integrating multi-modal and agentic capabilities directly on robots.

ContextNoiseCustomPrivacyModelsMaturity

Provides production-grade, robotics-specific SKILL.md prompts and working examples (QoS, lifecycle, bounded buffers, tests) to make coding agents generate production-ready ROS software.

ContextNoiseCustomPrivacyModelsMaturity

Provides an end-to-end, robotics-focused framework that benchmarks and improves Code-as-Policy agents with interactive robot gym environments, multi-tier benchmarks, a training-free agent, and RL fine-tuning tools bundled together.

ContextNoiseCustomPrivacyModelsMaturity

Integrates closed-loop AgentOS, 3D spatial memory, and embodied skills into a unified framework specifically designed for real-robot long-horizon tasks.

ContextNoiseCustomPrivacyModelsMaturity

Provides an auditable, fail-closed physical execution runtime that binds high-level agent intent to robot bodies and issues evidence-bearing receipts and leases.

ContextNoiseCustomPrivacyModelsMaturity

Provides an out-of-the-box robotics LLM stack with prebuilt movement, manipulation, vision, LiDAR and voice tools plus a GUI for running agents on real robots.

All repositories (16)

ROSA 🤖 is an AI Agent designed to interact with ROS1- and ROS2-based robotics systems using natural language…

1.6k
Python
1 yr

Connect AI models like Claude & GPT with robots using MCP and ROS.

1.4k
Python
1 yr

PhyAgentOS is a self-evolving embodied AI operating system built on agentic workflows.

1.0k
Python
4 mo

A Framework for Benchmarking and Improving Coding Agents for Robot Manipulation

661
Python
4 mo

RAI is a vendor agnostic agentic framework for Physical AI robotics, utilizing ROS 2 tools to perform complex…

555
Python
2 yrs

Agent skills that make AI coding assistants write production-grade robotics software. ROS1, ROS2, design patte…

322
+67d
Python
4 mo

A unified, agentic system for general-purpose robots, enabling multi-modal perception, mapping and localizatio…

305
Python
8 mo

Seamlessly integrate state-of-the-art transformer models into robotics stacks

294
Python
2 yrs

Self-evolving runtime infrastructure for Physical AI and embodied agents. Ground AI agents into robot bodies w…

170
Python
4 mo

🦾Make your robot autonomous with LLM agent. Set it up with the same ease as normal agents in CrewAI or Autoge…

124
Python
10 mo

Control robots and physical hardware with natural language through Strands Agents.

105
Python
5 mo

Advanced MCP Server ROS 2 bridging AI agents straight into robotics

83
Python
1 yr

A simple MCP server for the SO-ARM100 control

81
Python
1 yr

EmbodiedAgents is a fully-loaded ROS2 based framework for creating interactive physical agents that can unders…

65
Python
2 yrs

Offline fault diagnosis assistant for the AgiBot X2 Ultra humanoid: drop in a log package, get a three-round r…

22
Python
2 mo

A general-purpose robotic agent framework based on LLMs. The LLM can independently reason, plan, and execute a…

22
Python
7 mo

More in Applied agents