hAcKlyc/MyAgents

MyAgents – A Local‑First AI Agent Workbench

What it is – MyAgents is an open‑source desktop application that lets you create, run, and coordinate AI agents on your own computers. It provides a graphical workbench plus a command‑line interface so you can chat with agents, turn conversations into persistent tasks, and link multiple devices together in a private “AgentNet”. All data and execution stay on your machines; only optional cloud services (for device discovery and shared spaces) are used.

Key capabilities

Feature	How you use it
Multiple Agent runtimes	Built‑in support for Claude Agent SDK, DeepSeek Harness and Codex; you can also point to any model via API keys.
Task system	Create one‑off or recurring tasks (cron‑style) that run scripts, monitor conditions, and wake an agent when needed.
Channel IM integration	Connect agents to Feishu, DingTalk, Telegram (and extensible plugins) so you can interact from your favorite messenger.
AgentNet	Private network that lets agents on different computers (macOS 13+, Windows 10+) call each other, share files, and execute work in the appropriate environment.
Space collaboration	Shared “Space” with Issues, Goals, Skills and tools; you can invite collaborators or their agents to work on a common objective.
Desktop UI & CLI	Full‑featured GUI with workspace explorer, conversation history, file preview, and a lightweight floating assistant; the myagents CLI lets scripts or external agents create tasks, call other agents, and read results.
Skill & tool plugins	Define reusable Skills (methods) and connect to external tools via STDIO, HTTP or SSE.
Record & voice	Capture notes, audio, and local speech‑to‑text (requires a model that supports it) for later processing by agents.

Typical workflow

1. Install the client (download a macOS DMG or Windows installer – no Node.js or server needed).
2. Add a model in Settings → Model Providers – log in to a subscription or enter an API key.
3. Create a workspace (or use the built‑in “Mino” template) and spin up an agent with its own model, runtime, and Skills.
4. Chat with the agent, reference files (@file) or invoke Skills (/skill).
5. Turn a conversation into a Task – schedule it, set a cron, or enable “Goal Mode” for continuous work.
6. Connect other devices via AgentNet; assign a task to a Windows machine for verification, or to a always‑on server for long‑running jobs.
7. Collaborate in a Space: share Issues, assign them to agents, and keep all results in one place.

Installation & quick start

• Download the appropriate installer from the Releases page or the official website.
• Run the installer; the app launches with a welcome screen.
• Open Settings → Model Providers, log in to Claude, DeepSeek, etc., or paste an API URL.
• Click Create Agent, pick a workspace folder, and start a conversation.
• Use the side‑panel to create a Task, set a schedule, or open the AgentNet view to see other devices.

Who might benefit

• Power users who want AI assistance for research, coding, or content creation without sending data to the cloud.
• Developers who need a local LLM‑powered assistant that can run build scripts, run tests, or manipulate files.
• Teams that want a private, self‑hosted AI workflow where each member’s machine contributes resources.
• Researchers exploring multi‑agent orchestration, task automation, or human‑in‑the‑loop AI systems.

License – The client is released under the AGPL‑v3. It is free for personal and commercial use, but any network‑served modifications must also be made available under the same license. A commercial license is offered for those who need to embed or redistribute MyAgents without the AGPL obligations.

Where to learn more

• Official website & download page: https://myagents.io
• Full documentation (English & Chinese) inside the repo (README.en.md, bundled-skills/...)
• Architecture overview: specs/ARCHITECTURE.md
• Contribution guide: CONTRIBUTING.md

---

MyAgents aims to give every user a “super‑charged” personal assistant that runs locally, integrates with existing tools, and can be networked across their own devices.