What is Open Duet?
Open Duet is an open-source workspace for creating named AI agents with assigned roles, workspaces, memory, and approval settings. Agents can retain editable notes, search conversation history, and work together through delegation and shared rooms. A task board, calendar, and attention view help users follow work and review requests for approval.
The software can run as a desktop application or on a self-hosted server. A separately operated hosted service is also available. Open Duet does not include a model or model subscription: users connect their own command-line engine or API-based provider. Desktop and self-hosted instances store conversations, settings, and agent notes locally or in their server database, while selected engines receive the context required for each turn.
The project remains experimental. Self-hosting is intended for one owner with paired devices rather than a multi-user organization. Delegation, isolated agent computers, desktop installers, and several connected-service features have limited validation or engine-specific restrictions.
Key capabilities
- Agent coordination Create agents with individual roles and workspaces, and let Claude or Codex leads delegate tasks to specialists.
- Persistent context Save, recall, review, and edit agent notes, and search previous conversations.
- Scheduled routines Run prompts on a schedule, through webhooks, or manually while the backend remains active.
- Approval controls Select Ask first, Accept edits, or Plan only, and review tool activity and approval decisions.
- Tools and workspaces Connect selected agents to apps through Composio, add individual MCP tools, and work with workspace files.
- Work tracking Organize assignments in a task board and calendar, and review approvals, questions, failures, and routine outcomes.
Where Open Duet fits
Open Duet can fit into an individual's recurring research, project briefing, and file-based work. A lead agent can distribute preparation tasks to specialists and assemble their results, while routines repeat an established workflow. Users can review outputs and approve actions when required by their settings.
Before deployment, users should check engine compatibility, provider costs, workspace permissions, and which context reaches external providers. Scheduled work requires a running backend. Connected apps, isolated computers, and packaged desktop behavior need validation in the intended environment; the repository identifies several features that have not been qualified against live services or packaged applications.
Reviews of Open Duet
Be the first to share your experience with Open Duet.
Write a review
Loading review form…
