Pick the agent. Bring the model.
Use local models with Codex, Claude, Cursor, OpenCode, Grok, Hermes, and Pi. Keep the project, conversation, and review tools around them.
Your agents. Your machine.
Code here.
Go anywhere.
A home for your coding agents and local models. Build on your desktop. Pick up from your phone. Keep your flow.
Eight agents. One home port.
Plus your local models-
Codex -
Claude -
OpenCode -
Cursor -
Grok -
Hermes -
Pi -
Antigravity
A home port for everything you’re building.
Actual app. Sample projects and conversation.macOS / Windows / Linux
01 / Local by choice
Your models.
Your hardware.
Give your coding agents a local brain. Connect a model server you run, then choose its models in the same workspace you use for cloud agents.
-
LM Studio -
Ollama -
llama.cpp -
vLLM -
OpenAI-compatible
A local model. A familiar workflow.
One local setup.
Seven ways to build.
- 01
Connect your server
Start your local model server, then enable it in Settings → Local LLM.
- 02
See your models
Your server’s available models appear together in the Models tab.
- 03
Enable your harnesses
Toggle which supported agents can use your local models.
Mock settings demo · Sample server and models. Compatible server API required.
Compatibility details ↗More than one local server.
Connect multiple runtimes, choose which agents can use each one, and show the models you want in the picker.
See what’s loaded.
Inspect loaded LM Studio model instances and unload them when you need room for something else.
02 / The workspace
Less juggling.
More building.
Give every project, conversation, and code change a place to land. Everything you need to move the work forward, together.
A clear head for every project.
Keep your agent threads organized by project. Move between tasks with the conversation and working context right there.
See what actually changed.
Follow the plan, inspect changed files, and review the diff. Go from the agent’s answer to the code behind it.
Room to work in parallel.
Give separate tasks their own Git worktrees. Keep your branches and agent conversations connected as you build.
See the thing you’re building.
Open your app in the integrated desktop browser. Inspect pages, check responsive layouts, and keep a terminal close by.
Keep reviews moving.
Browse pull requests across GitHub, GitLab, Bitbucket, and Azure DevOps. Open a review with your agent, right from the inbox.
A workspace across machines.
Connect another Harbordex server or launch one over SSH from the desktop app. Work with the projects and agents on that machine.
03 / Beyond the desk
Step away.
Stay in the loop.
The coffee can wait.
Your next idea doesn’t have to.
Start a task, answer a question, or review a change from your phone. Your agents keep running on your machine, with your files and your tools.
- Continue the same conversations
- Review diffs and steer the next step
- Connect over Wi-Fi or Tailscale
iOS & Android · In development · Build from source
Your desktop, within reach.
04 / Your machine stays yours
Close to your code.
Even from a distance.
Your phone connects to the machine running your agents. Pair a device once, then get back to the work.
Your phoneWhere you are
Wi-Fi or TailscalePaired by you
Your machineWhere agents run
Direct on your network.
Use your local Wi-Fi or a private Tailscale network. On these connections, Harbordex doesn’t proxy traffic between your devices.
You decide who connects.
Pair with a QR code or pairing link. Each paired device is an authorized client you can revoke.
05 / Make yourself at home
A few steps.
A better way to work.
- 01
Choose your home base.
Install Harbordex on your desktop, or run the headless server on a remote machine.
- 02
Bring your agent.
Configure an agent in Settings → Providers. Sign in to its service, or connect a compatible local model server, then start a thread.
- 03
Take it with you.
Build the mobile companion and pair your phone from Settings → Connections.
The desktop app
For your daily driverbrew install --cask harbordex winget install Harbordex.Harbordex yay -S harbordex-bin The headless server
For a remote machinenpx harbordex serve npx harbordex serve --host "$(tailscale ip -4)" npx harbordex serve --tailscale-serve Pair with the QR code in your terminal.
What is Harbordex?
Harbordex is a desktop workspace for coding agents, with a mobile companion. It brings projects, conversations, terminals, and code review together. It is a fork of T3 Code and is in early development.
Which agents can I use?
Codex, Claude, OpenCode, Cursor, Grok, Hermes, Pi, Antigravity. Install or configure the agent you want in Settings → Providers; sign-in and model requirements vary by agent. Gemini, GitHub Copilot, and ACP Registry are listed as coming soon in the app.
Can I use local LLMs?
Yes. Connect LM Studio, Ollama, llama.cpp, vLLM, or an OpenAI-compatible server in Settings → Local LLM. Choose a tool-capable model and enable it for a supported agent. Codex, Claude, Cursor, OpenCode, Grok, Hermes, and Pi can use local models; Antigravity currently cannot. Compatibility depends on the agent and the server’s API: Codex uses Responses, and Claude requires Anthropic Messages support.
Do I need another AI subscription?
Harbordex is free to run. Use an existing provider subscription or API account, or connect a compatible local model server. Hosted providers’ pricing and usage limits still apply; local inference uses the hardware running your model server.
Does my code run in the cloud?
Your agents run on the desktop or remote machine you connect. Your selected model endpoint handles inference: a hosted provider or a local model server you configure. Tools used by an agent can still access external services. Harbordex does not move your workspace into a hosted sandbox.
How do I get the mobile app?
The iOS and Android companion is in development and is not yet distributed through app stores. You can build it from source, then pair it with your desktop or headless server.
Do I need a Harbordex account?
No account is required for the desktop workspace and direct device pairing. Configure your agents and model endpoints locally and authorize the devices you want to connect.
A home for your next big idea
Good work
travels with you.
Get Harbordex macOS, Windows & Linux. Bring your favorite agent.