Skip to main content

Agent Operating System — Production-ready multi-model agent framework with Tool-Using Agent, LLM Provider abstraction (5 providers), Function Calling, streaming, retry, checkpoint/resume, swarm coordination, self-evolution autopilot, hybrid search (BM25+dense+rerank), agent evaluation suite (SWE-bench), skill ecosystem bridge, A2A protocol, comprehensive observability, and enterprise features (RBAC, multi-tenancy, audit logging, API key management).

Project description

AgentOS - Universal Agent Runtime

PyPI version License: MIT Python 3.10+

AgentOS is an open-source universal agent runtime designed for the multi-agent era. It provides a unified execution environment where AI agents can discover, install, and compose skills from a decentralized marketplace, while connecting to any MCP-compatible tool server.

pip install nexus-agentos

Why AgentOS?

Problem AgentOS Solution
Agents siloed in different frameworks Universal runtime — one env runs OpenClaw, LangChain, custom agents
No marketplace for agent skills Built-in Skill Market — devs upload, users install, admins review
MCP servers fragmented 8 built-in MCP servers, 34 tools out of the box
No TUI for agent management Terminal UI (Textual) — file browser, skill manager, task runner
Hard to compose multi-agent workflows Sub-agent dispatch — agents delegate to specialists automatically

Quick Start

1. Install

pip install nexus-agentos

# Or from source
git clone https://github.com/wuyifeishu/nexus-agentos.git
cd nexus-agentos && pip install -e .

2. Launch the Skill Marketplace

python -m agentos.server.marketplace_platform

Open http://localhost:8899/static/platform.html — you'll see the full marketplace with:

  • Browse: search/discover 64 built-in skills across 11 categories
  • Login/Register: JWT auth for developers
  • Upload: zip your skill with a skill.yaml manifest
  • Review: admin panel for security scanning, approve/reject

3. Run the Terminal UI

python -m agentos.desktop

Ctrl+M opens the Market Panel. Ctrl+F opens File Browser. Full keyboard-driven.

Architecture

┌─────────────────────────────────────────────────────────┐
│                    AgentOS Runtime                       │
│  ┌─────────┐  ┌──────────┐  ┌───────────────────────┐  │
│  │   TUI   │  │  Agent   │  │   Sub‑Agent Dispatcher │  │
│  │(Textual)│  │  Engine  │  │  (search/browser/file) │  │
│  └─────────┘  └──────────┘  └───────────────────────┘  │
│  ┌─────────┐  ┌──────────┐  ┌───────────────────────┐  │
│  │  Skill  │  │   MCP    │  │   Marketplace Server   │  │
│  │Registry │  │ Registry │  │  (FastAPI + SQLite)    │  │
│  └─────────┘  └──────────┘  └───────────────────────┘  │
│  64 Skills  │  8 Servers   │  Upload/Review/Download  │
└─────────────────────────────────────────────────────────┘

Built-in MCP Servers

Server Tools Description
filesystem 7 read_file, write_file, list_directory, search_files, get_file_info, create_directory, move_file
webfetch 3 fetch_url, fetch_json, check_url
memory 6 store_memory, retrieve_memory, search_memory, list_categories, delete_memory, update_memory
search 4 web_search, news_search, image_search, suggest
git 4 git_status, git_log, git_diff, git_branch
shell 3 run_command, system_info, disk_usage
code 3 run_python, run_shell, lint_code
text 4 count_tokens, extract_regex, summarize_text, format_json

All MCP servers are zero-dependency, pure Python implementations. No external processes required.

Skill Categories (64 built-in)

Category Count Examples
Development 14 github, docker, git, kubernetes, terraform, postgres, redis
Productivity 8 google-workspace, microsoft-365, calendar, email, todoist
Communication 7 slack, discord, notion, jira, teams, webex, zendesk
Data 8 sql-query, csv-tools, bigquery, snowflake, mongodb, elasticsearch
Document 5 docx, xlsx, pptx, pdf, markdown
Media 5 image-edit, video-process, audio-transcribe, screen-capture, svg-create
System 4 system-info, file-watch, cron-scheduler, network-tools
Security 4 secret-manager, threat-scanner, 1password, auth-proxy
AI 5 prompt-optimizer, token-counter, embedding-service, langchain-tools, rag-pipeline
Uncategorized 4 weather, translation, currency-converter, timezone

Skill Marketplace Platform

A production-ready marketplace for the AgentOS ecosystem:

For Developers

  • Register/login with email & password (bcrypt + JWT)
  • Upload .zip skill packages with skill.yaml/manifest.json
  • Automatic security scanning on upload: dangerous imports, shell injection, obfuscation, hardcoded keys, excessive permissions
  • Track downloads and reviews
  • Developer profile pages

For Users

  • Browse/search skills by name, category, tags
  • One-click install from any ecosystem source
  • Rate and review installed skills

For Admins

  • Review queue with security findings per skill
  • Approve/reject with comments
  • Skill quality and safety enforcement

API Endpoints

POST   /api/auth/register          - Create developer account
POST   /api/auth/login             - Login, receive JWT
GET    /api/skills                 - Browse/search published skills
GET    /api/skills/{id}            - Skill detail
GET    /api/skills/{id}/download   - Download skill zip
POST   /api/skills/upload          - Upload new skill (auth required)
GET    /api/my/skills              - My uploaded skills
GET    /api/developers/{username}  - Developer profile
POST   /api/admin/review/{id}      - Approve/reject (admin)
GET    /api/admin/review-queue     - Pending reviews (admin)
GET    /api/categories             - Category listing

Security Scanning

Every uploaded skill is automatically scanned for:

Check Severity Description
Dangerous imports Critical os.system, subprocess, socket, ctypes
Shell injection Critical os.system(), eval(), exec(), __import__()
Code obfuscation High Base64-encoded payloads, eval chains
Hardcoded secrets High API keys, tokens, passwords in source
File permission escalation Medium chmod 777, os.chown
Network exfiltration High Suspicious requests.post to unknown hosts

Roadmap

  • Universal Agent Runtime
  • 64 Built-in Skills
  • 8 MCP Servers (34 tools)
  • Skill Marketplace with web UI
  • Developer registration & upload
  • Security scanning pipeline
  • Admin review queue
  • Terminal UI (TUI)
  • Multi-agent orchestration dashboard
  • Skill dependency resolution
  • Federated marketplace discovery
  • Blockchain-based skill provenance

Contributing

We welcome contributions! The Skill Marketplace is our highest-priority area.

  1. Fork the repo
  2. Create a feature branch: git checkout -b feat/amazing-skill
  3. Write your skill with a skill.yaml manifest
  4. Upload to the marketplace or submit a PR
  5. If you want to contribute to the platform itself, check the issues labeled good first issue

Skill Manifest Format

name: my-awesome-skill
version: 0.1.0
description: Does something awesome
author:
  name: Your Name
  url: https://github.com/yourname
category: dev
tags: [python, automation]
entrypoint: main.py
requires:
  python: ">=3.10"
  packages: [requests, httpx]
license: MIT

License

MIT License. See LICENSE for details.

Star History

Star History Chart


Built with ❤️ by the AgentOS community. Let's make the Agent OS ecosystem thrive.

Project details


Release history Release notifications | RSS feed

This version

1.9.1

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

nexus_agentos-1.9.1.tar.gz (555.1 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

nexus_agentos-1.9.1-py3-none-any.whl (630.9 kB view details)

Uploaded Python 3

File details

Details for the file nexus_agentos-1.9.1.tar.gz.

File metadata

  • Download URL: nexus_agentos-1.9.1.tar.gz
  • Upload date:
  • Size: 555.1 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.11.6

File hashes

Hashes for nexus_agentos-1.9.1.tar.gz
Algorithm Hash digest
SHA256 0e1ef742d763a56c3f4c99eab1395be5e4b96518f71b6f37dad8b84792784aa0
MD5 5552abf6b4b6cd2dc83a79627f120c78
BLAKE2b-256 2bb95270115b5dbac2ee1910721f535d823524e74f5bbd0acf1c749ca9cde8e0

See more details on using hashes here.

File details

Details for the file nexus_agentos-1.9.1-py3-none-any.whl.

File metadata

  • Download URL: nexus_agentos-1.9.1-py3-none-any.whl
  • Upload date:
  • Size: 630.9 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.11.6

File hashes

Hashes for nexus_agentos-1.9.1-py3-none-any.whl
Algorithm Hash digest
SHA256 443b81e891d6633a63b68e22a250a5abac2010a49ea62732a0c49706442e0d59
MD5 4754eef6efbea8178bda4e0a87727727
BLAKE2b-256 faec27f397b91c611250c15d7ff75d54be83f35f705383f769940a72ea67ce2d

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page