Harness Studio
Find, run and connect AI models on your own computer. Harness Studio is an open-source desktop app. Pick a model, run it locally, and connect it to the tools you already use through an optional local gateway.
0.2.0, early public release
One app, two ways to work.
Start in Simple mode. Switch to Advanced mode when you want control over servers and performance.
Find your model. Run it. Make it part of your day.
The everyday path, with a few screens and clear next steps.
- Dashboard
- Find a Model
- My Models
- Chat and Media
- Connect an App
Manage your servers, tune your setup and explore performance.
The full workbench for people who run several servers and care about detail.
- Servers
- Create
- Performance
- Analytics
Built for how you work.
Choose a group to see what Harness Studio does for it.
Everything you need to chat with a local model.
- Find models that fit
- Search Hugging Face with hardware fit guidance. Downloads are SHA-256 verified.
- Playground
- Chat, save conversations and compare models. Use vision, local document search (RAG), speech and image generation.
- Safe tools
- Built-in tools ask for approval before risky actions.
- Harness Link
- Connect to another Harness instance.
- Made for you
- English and Turkish interface, light and dark themes, accessibility options.
A local endpoint your tools already understand.
- Compatible APIs
- An opt-in local gateway with OpenAI-, Anthropic- and Ollama-compatible APIs.
- Guided setup
- Step-by-step setup for Claude Code, Codex, Aider, VS Code and Open WebUI.
- Tool calls and MCP
- Tool-call handling for local models, and MCP in both directions.
- Custom tools
- Write your own JavaScript tools. They run in a sandbox.
Control, measure and automate.
- Servers and profiles
- Manage multiple servers with profiles and detailed llama.cpp controls.
- Performance
- Benchmarks, request analytics, Prometheus metrics and optional OpenTelemetry.
- CLI and Control API
- Script everything, or run headless with harness serve.
- More engines
- whisper.cpp, sherpa-onnx, ONNX Runtime, stable-diffusion.cpp, MLX and mistral.rs.
Connect the tools you already use.
Turn on the local gateway when you want other apps to talk to your models.
Gateway APIs
- OpenAI-compatible
- For tools that expect the OpenAI API.
- Anthropic-compatible
- For tools that expect the Anthropic API.
- Ollama-compatible
- For tools that expect Ollama.
The gateway is opt-in. It stays off until you turn it on.
Guided setup
Harness Studio walks you through each connection.
- Claude Code
- Codex
- Aider
- VS Code
- Open WebUI
MCP in both directions
Connect to MCP servers, and expose Harness tools through MCP.
Engines
Harness Studio manages the engines for you.
- llama.cppText models. Managed by default.Default
- whisper.cppSpeech recognition.
- sherpa-onnxSpeech with ONNX models.
- ONNX RuntimeONNX model inference.
- stable-diffusion.cppImage generation.
- MLXApple Silicon.
- mistral.rsRust inference engine.
Platform requirements
- Windows
- Windows 10/11, 64-bit.
- macOS
- macOS 13.3 or later. Apple Silicon runs with Metal. Intel is CPU-only.
- Linux (preview)
- Linux x86_64. .deb and AppImage packages.
Private by default.
Use Harness Studio as a guest, even offline. Downloading and running it does not require an account.
- No account needed to download or use the desktop app.
- Models run on your computer and keep their own licenses.
- Local documents and hardware details are not uploaded by default.
Questions
Do I need an account?
No. You can download and use Harness Studio as a guest, including offline.
Is it free and open source?
Yes. Harness Studio is free to use and open source under the MIT license. Models you download keep their own licenses.
Which systems does it run on?
Windows 10/11 (64-bit), macOS 13.3 or later, and Linux x86_64 as a preview with .deb and AppImage packages.
Which engine does it use?
It manages llama.cpp by default. whisper.cpp, sherpa-onnx, ONNX Runtime, stable-diffusion.cpp, MLX and mistral.rs are also supported.
Can I use it with Claude Code or VS Code?
Yes. Turn on the opt-in local gateway and follow the guided setup for Claude Code, Codex, Aider, VS Code or Open WebUI.
Which languages does the app speak?
The interface is available in English and Turkish.
Start on your own computer.
Download Harness Studio without an account, or read how to install it.