agentlas-ai/Agentlas-OS

Agent OS: keep specialist agents in a hub, spin up a temporary orchestrator per task. Local-first, works with any model.

What it solves

Agentlas OS solves the problem of efficiently routing complex tasks to the most qualified specialist agents without relying on biased rankings. It prevents the "one-size-fits-all" approach where a single model with different prompts handles everything, instead staffing a team of domain-specific agents (from local, cloud, or public sources) to complete a job with minimal token usage.

How it works

The system operates in three primary stages:

  1. Task Split: A high-level request is broken down into specific tasks, defining roles, hand-off order, and required artifacts. It automatically rejects circular dependencies in the workflow.
  2. Agent Staffing: The system identifies candidate agents from the user's machine, private cloud, and the public Agentlas Hub. Instead of ranking them, it provides a list of candidates with verified identities (via signed agent cards) and qualification levels, allowing the user's model to select the best fit.
  3. Run and Verify: Planning, execution, and synthesis are run as separate invocations. An independent verifier must approve the work before it is marked as complete, with every step proven by a receipt.

Who it’s for

It is designed for users of AI coding assistants and CLI hosts (such as Claude Code, Cursor, Codex, and Gemini) who need specialized expertise across various domains, ranging from software engineering (backend, security, QA) to professional services like insurance actuary or M&A diligence.

Highlights

  • Standardized Protocol: Uses A2A v1.0 and /.well-known/agent-card.json for agent identity and capability verification.
  • Token Efficiency: Compresses candidate lists to reduce token costs while maintaining the probability of finding the right specialist.
  • Cross-Host Integration: Configures multiple agent hosts on a single machine simultaneously, including MCP server registration for Cursor and Copilot.
  • Model Fallback: Automatically re-runs a turn on the next model in a priority list if the current model hits a quota limit.
  • Agentlas Hub: A free community for publishing and discovering public agents.

Related

  • Project
    ag2ai/ag2

    An open-source AgentOS framework for building AI agents and orchestrating their cooperation through a hub-and-channel network system.

  • Project
    777genius/agent-teams-ai

    Agent Teams AI is a free desktop app that lets you assemble, run, and monitor teams of AI agents across many LLM providers. Agents collaborate on a Kanban board, can review code, and you retain control via per‑action approvals and budget analytics. Installable on macOS, Windows, and Linux with no build required.

  • Project
    Orkas-AI/Orkas

    Orkas is a real, substantive project: an open‑source (MIT) local‑first multi‑agent desktop application. It uses a central “Commander” agent that breaks goals into to‑dos, delegates to 16 built‑in specialist agents, and can orchestrate external CLI coding/personal agents (Claude Code, Codex, OpenCode, OpenClaw, Hermes‑Agent). It includes projects with persistent memory, auto‑advancing to‑dos, scheduled automations, document retrieval (RAG) via a local SQLite vector index, and agents that can learn from their own activity. Built with Electron/Node, it runs locally and supports many AI providers.

  • Project
    atlassdk/atlas

    Atlas is a decentralized coordination layer that allows AI agents across different blockchain ecosystems to discover each other, negotiate tasks, and settle payments.

  • Project
    code-yeongyu/oh-my-openagent

    An AI agent command-line tool that automates complex professional workflows like coding and deep research using multi-agent graphs and a persistent memory system.