Agentlas-OS
The Open-Source AI Agent OS inspired by OpenAI Dots, Meta Muse, OpenClaw, Hermes & Mem0. Autonomous 3-tier headhunting, Receipt & Verifier loops, scoped memory,
What is Agentlas-OS?
The Open-Source AI Agent OS inspired by OpenAI Dots, Meta Muse, OpenClaw, Hermes & Mem0. Autonomous 3-tier headhunting, Receipt & Verifier loops, scoped memory, and BYOM for Claude Code, Codex & Cursor.
Overview of Agentlas-OS
What it is
Agentlas OS is an open-source AI agent operating system inspired by OpenAI Dots, Meta Muse, Claude Duet, Nous Hermes, and Mem0. It is designed to lead entire projects rather than isolated prompts. It is model-agnostic (BYOM) and works natively with Claude Code, Codex, Cursor, Antigravity, and Agentlas Desktop. Unlike other agent products that run in virtual machines or use prompt loops without verification, Agentlas OS runs no VM. Agents work directly in the terminal, files, and tools already on the user's machine. A task is considered complete only after an independent verifier accepts its execution receipt. The engine is open source under Apache-2.0. As of the README, there are over 12,000 desktop downloads, and it is free to download, running on the user's own AI plan or API key.
Key features
- Persistent autonomous execution: Goals survive across sessions, work resumes from the last checkpoint, and each step is verified by an independent verifier.
- Domain-complete workflows: Includes One (personal agent), Finance (investment research with traceable calculations), and Science (molecular and literature research).
- Desktop execution, mobile decisions: Execution stays on the user's computer; decision points are sent as question cards to the phone, and the desktop continues after the user answers.
- Portable skills and A2A packaging: The
40-session-agent-buildercompiles procedures and checks into a package;30-agentlas-packagerpackages existing agents. Each package carries an A2A Agent Card and installs on any supported host. - Scoped memory governance: Memory hooks, context maps, and project indexes are scoped per project. A Memory Curator validates, redacts, and deduplicates memory events; a Policy Gate approves promotion to shared team memory.
- Hephaestus engine: Decomposes a goal into a task graph, recruits specialists from three tiers (local machine, private Agent Cloud, public Hub), and verifies completion via independent verifier.
- Bring your own model: Supports Claude Code, Codex, Antigravity, Grok, Kimi, Cursor, GitHub Copilot CLI, or local models via Ollama, LM Studio, or MLX.
Use cases
- Personal agent (One): Runs on the user's computer, opens work sessions, assigns agents, sends question cards, and attaches results as files.
- Investment research (Finance): Save research criteria as an Analyst; reports keep the inputs behind each calculation.
- Scientific research (Science): Molecular views and typeset manuscripts in one app, as a signed Agentlas Desktop extension.
- General project leadership: Decompose complex objectives, recruit specialist agents, and verify each step before marking work complete.
Getting started
Install the runtime using the provided script: bash curl -fsSL https://raw.githubusercontent.com/agentlas-ai/Agentlas-OS/main/scripts/install-all-runtimes.sh | bash After installation, run: bash ~/.agentlas/os-bin/agentlas-os features status Alternatively, paste the installation request into Claude Code, Codex, Antigravity, or Cursor as described in the README. Agentlas Desktop is a separate application with its own installation and update flow.
License and pricing
Licensed under Apache-2.0. Open source and free to download. Users run it on their own AI plan or API key.
Based on the project's public README and repository metadata.