OpenAgent

A native AI workspace

Give capable agents
a place to work.

Run capable agents with project context, tools, memory, skills, and reversible changes in one native desktop workspace.

Download OpenAgent View on GitHub

Stable and Beta releases

Download for your computer.

Release notes
WindowsWindows 10 / 11x64 installer · .exe macOSApple SiliconM1, M2, M3, M4 · .dmg macOSIntel MacIntel processor · .dmg LinuxDebian / Ubuntux64 package · .deb LinuxFedora / RHELx64 package · .rpm LinuxUniversal Linuxx64 portable app · AppImage

Select Stable or Beta to see its latest published build.

Native
A fast Tauri + Rust desktop app
Inspectable
See tool calls, approvals, diffs, and history
Extensible
Bring your own models, MCP tools, and skills

One workspace, the complete loop

More than a chat window.

OpenAgent combines an agent runtime, project tools, durable context, and desktop controls so useful work can continue across turns and sessions.

01

Agent runtime

Stream primary conversations, delegate nested sub-agents, ask structured follow-up questions, or run autonomous Goal and Graph workflows.

02

Tools & MCP

Work with files, terminal commands, fetched pages, search, and external MCP servers connected over HTTP or stdio.

03

Memory & skills

Recall useful knowledge across sessions with hybrid search, and load reusable global or workspace-scoped SKILL.md instructions.

04

Checkpoints & branches

Search conversations, keep a durable checkpoint for every turn, branch from earlier messages, inspect diffs, and roll work back.

05

Rich, interactive output

Render streaming Markdown, code, Mermaid, ECharts, interactive file and URL chips, plus sandboxed HTML previews with PNG export.

06

Models & context

Connect Anthropic, OpenAI, or compatible providers, switch models, attach multimodal files, and compact long conversations.

07

Scheduled tasks

Run recurring or one-off chat hooks in the background, restore them on startup, and receive desktop notifications.

08

Workspace knowledge

Keep project drafts, memory, skills, and DESIGN.md guidance separated by workspace or shared globally when needed.

09

Desktop experience

Use light or dark themes, Chinese or English, system tray controls, launch at startup, and native file-manager actions.

Control without losing momentum

Autonomy,
with receipts.

Choose how independently agents can act while preserving the evidence needed to understand, approve, or undo their work.

01

Flexible approvals

Review every call, use model-assisted approval, disable approval, or apply the workspace sandbox.

02

Workspace boundaries

File and terminal actions stay aligned with the active project and its configured access policy.

03

Durable recovery

Conversation snapshots, branches, cancellations, and reverse diffs make experimentation recoverable.

04

Optional observability

Inspect local run details and optionally connect Langfuse tracing when deeper diagnostics are useful.

A clearer way to work

Context that
doesn't disappear.

  1. 01

    Connect your model

    Add an Anthropic, OpenAI, or compatible provider and choose the model that fits the task.

  2. 02

    Open the right workspace

    Select a project, attach relevant files, and let local memory, skills, and DESIGN.md supply focused context.

  3. 03

    Watch and steer the work

    Follow streamed reasoning artifacts, tool calls, approvals, sub-agents, graph progress, and file changes as they happen.

  4. 04

    Keep or reverse the result

    Review the diff, retain useful knowledge in memory or skills, branch to explore another path, or roll changes back.

Open source desktop AI

Make your next
agent task feel grounded.

Download the latest desktop release, connect a provider, and build a workspace that works the way you do.

Download the latest release