OpenAgentd

Open-source desktop app for a local coding agent. Watch every tool call, diff, and terminal command. Bring your own models; sessions stay on your machine.

Automatically checked · GitHub checked 2026-11-12

OpenAgentd

Overview

OpenAgentd: the coding agent that shows its work

Latest release Downloads Apache 2.0 Discord

Website · Install · Tour · Features · Discord · Contribute

OpenAgentd is a desktop app for running a coding agent on your own machine. You give it a folder and a task. It plans, reads, edits, runs your tests, and shows you every step: each tool call with its arguments and timing, every diff, and what each turn costs. You can steer it mid-turn, review its changes before you commit, and use the models and keys you already have.

OpenAgentd workspace: the agent's tool calls and test runs on the left, its uncommitted Git diff on the right

A real session, unedited: the agent added rate limiting to a small Bun service, ran the suite, and left the changes uncommitted for review.

Why OpenAgentd

  • Nothing happens off-screen. Tool calls, reasoning, shell output, file edits, todos, tokens and cost stream into one transcript. Collapse it to just the answers with Reader mode when you only want the result.
  • You decide when code changes. Plan mode explores read-only and writes a plan you can comment on, then approve. Approve with Keep going and the agent works until the plan's Done when is proven. In Code mode you can steer, queue, or stop a running turn.
  • Made for real repositories. Git changes, history and graph, worktrees, a real terminal, a live preview of your web app, and language-server diagnostics after edits.
  • Your models, your keys. 18 built-in providers: Anthropic, OpenAI, Gemini, OpenRouter, Bedrock, DeepSeek and more, your existing ChatGPT, Copilot or Grok subscription, and local models through Ollama. Plugins can add others. When a task needs an API key, the agent asks for it on a card and uses it without ever seeing it.
  • On your phone too. The iPhone and Android apps show the same sessions: answer the agent's questions, approve plans and follow its work, over your network or from anywhere with optional remote control.
  • Extend it. Connect MCP servers, add skills and plugins, generate images and video, and let the agent chain many tool calls in one scripted step.
  • Local-first, and small. A single native Rust backend: a 46 MB binary that starts in about 50 ms. Sessions, settings and telemetry stay on your machine. Other devices connect only when you turn on network access or sign in to remote control.

Install

macOS and Linux

curl -fsSL https://openagentd.dev/install.sh | sh

Windows (PowerShell)

irm https://openagentd.dev/install.ps1 | iex

Homebrew (macOS)

brew install --cask lthoangg/tap/openagentd

Open OpenAgentd and connect a model on the first screen: paste an API key, sign in with a ChatGPT or Copilot subscription, or choose Use on a key or Ollama install it finds. Then open a folder and describe what you want built. The desktop app starts its own backend; there is nothing else to run. The setup guide walks through providers and first launch, and builds for every platform are on the releases page. Unsigned builds may ask you to confirm before the first launch.

iPhone and Android apps connect to OpenAgentd on your computer. The iPhone and iPad app is on the App Store, with early builds in the public TestFlight beta; see the iOS and Android guides.

To build from source, see CONTRIBUTING.md.

CLI server

Install only the backend to run it from a terminal or to serve the apps on your other devices:

curl -fsSL https://openagentd.dev/install.sh | sh -s -- --cli
agentd server start                       # add --public --key --qr to pair a phone, or --cloud for remote control
agentd run "Summarize this project"
agentd run --continue "Now add tests for it"
# Windows PowerShell
& ([scriptblock]::Create((irm https://openagentd.dev/install.ps1))) -Cli

The server serves the API only: connect to the printed Server: address from the desktop or mobile app. There is no built-in TLS, so put an HTTPS reverse proxy in front of it before exposing it outside a trusted network. agentd --help lists every command.

A quick tour

Plan review with comments on the plan and Approve

Plan first, then build

In Plan mode the agent explores without editing and writes a plan. Select any passage to comment on it, then Approve, or Reply in chat to send your comments as a normal message. Approve with Keep going and it works until the plan is done. Press Tab in the composer to switch modes.
A lead agent summarising an explorer subagent's report

Hand work to subagents

The lead agent delegates focused jobs to subagents such as explorer#1. Their reports fold into the conversation, and the sidebar shows who is working on what.
A side question answered beside the running session

Ask on the side

Type /btw to ask about the work without interrupting it. The answer opens in its own panel and never enters the agent's context.
The same session on a phone

Take it with you

Pair a phone by scanning a QR code. It shows the same sessions, rings when the agent needs you, and lets you reply, approve or stop from anywhere.

The full tour covers agent teams, goals, the scheduler, web preview, secrets, plugins, MCP and skills, memory, telemetry and more. The feature catalogue lists every shipped capability with the release that introduced it.

Community

OpenAgentd is open source under the Apache 2.0 license.