Suparagent: Open-Source AI Coworker Workspace

What changed

Suparagent is an open-source project designed as a self-hosted alternative to proprietary AI coworker platforms like Grok Bot, Claude Duet, and OpenAI Dots. It allows users to create named AI agents, assign them specific jobs and workspaces, and coordinate their efforts to complete tasks. The platform centralizes conversations, memory, routines, and approval mechanisms within a single environment. Suparagent supports both a desktop application and a self-hosted server option, with a separate hosted service also available. The project is in active development, with features like desktop releases, self-hosting capabilities, agent computers, and team functionalities still evolving. The latest release mentioned is v0.8.5.

Key functionalities include:

• Agent Creation and Management: Users can create agents with names, roles, engines, workspaces, memory settings, and approval modes. Lead agents can delegate tasks to specialist agents.
• Context and Memory: Agents can save notes, recall relevant information from conversations and history, and search their past interactions. Users can review and edit agent memories.
• Routines and Scheduling: Tasks can be scheduled to run automatically, via webhooks, or on demand. Routines require explicit approval for actions, preventing autonomous execution.
• Approval Workflows: Users can configure agent freedom through modes like "Ask first," "Accept edits," "Approve for me," or "Plan only," with tool activity and approval decisions visible in the app.
• Tool and App Integration: Suparagent supports connecting services like GitHub, Gmail, and Google Calendar via Composio keys, granting specific agents access, with changes awaiting user approval. Individual tools can be added from the MCP Marketplace.
• Unified Dashboard: A "Today" view consolidates approvals, questions, routine outcomes, engine failures, and due tasks, with optional read-only Gmail and calendar items.
• Task Management: A task board and calendar facilitate work assignment, scheduling, and review of agent conversations and attempts.
• Instruction Reuse: Skills and templates allow for saving and reusing agent instructions.
• Mobile Access: A phone app pairs with a self-hosted server for chat, action approval, and agent/room management.
• File and History Management: Users can search conversations, attach files, import audio for transcription, browse workspace files, and export history.

Suparagent offers multiple ways to run: a desktop app (experimental, requires Node.js 22+, uv, and an engine), a self-hosted server (experimental, requires Linux server, HTTPS, Docker/Node.js+uv, and an engine), a hosted web app (separate product), and a development checkout. The project does not include a model or model subscription; users must bring their own engine (e.g., Claude Code/Codex CLI, Gemini CLI, Google Gemini via Antigravity, Cursor, Ollama, LM Studio) and pay for model usage.

Why it matters for builders

Suparagent provides a robust, open-source framework for building sophisticated AI agent systems. Its self-hosted nature and support for various AI engines empower developers to create custom agent workflows without vendor lock-in. The granular control over agent autonomy, memory, and approval processes allows for the development of reliable and secure AI coworkers tailored to specific business needs.

Practical impact

Developers can explore Suparagent by cloning the GitHub repository and running the desktop app using Node.js. They can configure their preferred AI engine and start creating agents with defined roles and workspaces. For more advanced use cases, self-hosting the server offers persistent agent availability. Builders should review the "Known limitations" section, particularly regarding the experimental status of the self-hosted server and mobile app, and the specific engine requirements for team delegation.

Caveats and source limits

The source material indicates that Suparagent is in active development, with several features, including the self-hosted server, team delegation, and the phone app, being experimental or having limitations. The desktop installers are unsigned, and the release-to-release update path has not been fully verified. The self-hosted server is intended for single-owner use, not multi-user organizations. The phone app has not been tested on iOS and lacks push notifications. Docker computers, browser tools, and the macOS native desktop preview are experimental and opt-in. Pricing for the hosted plans is provided, but the specific costs for using self-hosted engines are dependent on the chosen AI provider.