AI Agents
agent-qa
A self-improving QA agent that writes tests in plain English.
★ 902 starsNOASSERTION
View guide →Recursive self-improving physical agent OS — verified, evidence-based execution for robots and simulators
PhyAgentOS-core is an open-source ai agents project by PhyAgentOS (PhyAgentOS/PhyAgentOS-core on GitHub). It has 2,776 GitHub stars and 128 forks, is released under the MIT license, and was last updated on GitHub on 8 October 2026.
Define an embodied task with a visible goal and success criteria — PAOS verification modes (audit, enforce, recovery) are built around explicit criteria.
Clone the repo and install it in editable mode; install Dora CLI v0.4.1 only if you will run managed Forge Skill profiles.
Run paos onboard, then edit ~/.PhyAgentOS/config.json with your model provider (e.g. OpenRouter) and Forge/verification settings, and install the Forge Skills you need.
Start an installed Skill Runtime, then paos agent -m "..." with a task, and check the verdict and evidence afterwards.
Robot actions enter only through the versioned Forge Gateway Tool API (Query/Action/Session) — the agent never reaches directly into a simulator or hardware SDK.
Validated images and optional robot state are captured around actions with source, sequence, time and SHA-256 digest metadata.
The verifier judges goal, criteria, constraints, execution facts and evidence — with off, audit, enforce and recovery modes.
A recovery verdict appends a bounded PlanRevision to the same task; unknown effects are reconciled, never blindly retried.
SQLite WAL transactions persist AgentTask, PlanRevision, Query records and Gateway invocation references.
Verified tasks produce reusable workflow Skills and clustered Lessons, loaded only with the matching Skill, with guarded promotion.
Clone PhyAgentOS-core and install it editable, exactly as the 5-minute quick start shows.
git clone https://github.com/PhyAgentOS/PhyAgentOS-core.git
cd PhyAgentOS-coreOnboard to create ~/.PhyAgentOS/config.json and the default workspace, then add your provider API key to the config.
paos onboardRun the interactive CLI (or paos gateway for channels/Cron/Heartbeat) and give it a verifiable task.
paos agent
# one-shot: paos agent -m "Inspect Forge capabilities, then place the object in the target area and verify the visible result."The PhyAgentOS project team maintains it; the project's site is phy-agent-os.x-era.com.
Yes, it is MIT-licensed. You supply your own model provider and any robot/simulator hardware or software.
The config example uses a provider such as OpenRouter with your API key, so model costs depend on your provider and model choice.
General frameworks plan and call tools. PAOS adds a single governed execution gateway for physical actions, captured evidence, a separate verifier, and experience that only comes from verified tasks.
pip install -e . from a clone, run paos onboard, configure a provider in config.json, install a Forge Skill, and run paos agent.
A self-improving QA agent that writes tests in plain English.
A self-improving agent for coding workflows and long-running autonomous tasks
Self-improving memory layer that works across AI agent harnesses.
AI agent framework for plan-first development workflows with approval-based execution. Multi-language support (TypeScript, Python, Go, Rust) with automatic t…
WebArena-Verified: the audited, deterministically scored release of the WebArena web-agent benchmark.
An open-source multimodal retrieval engine for agent pipelines.
The most-starred other AI Agents projects in the LibreAI catalogue — the realistic shortlist if PhyAgentOS-core is not the right fit.
A personal AI that really does things — on any OS, with your data staying yours.
An agentic skills framework & software development methodology that works.
Agent harness optimization system with skills, memory and security for Claude Code, Codex and more
The agent that grows with you — a Python framework for adaptive AI agents.
DeepSeek's plugin-based agent harness where everything is a plugin.
Free and open — grab the source, star the repo, and start creating.