Overview
Shun is a local-first coding harness designed for running coding agents on hardware that the user owns. It is open source under the MIT license and focuses on small language models, such as a 7B model, by managing context carefully and providing bounded tool outputs. The product runs on the user's machine without telemetry, accounts, or a server operated by the developers, keeping code, transcripts, and files local by construction. It supports optional cloud providers only when explicitly chosen per task.
The harness includes features like atomic edits, validated tool arguments, provider adapters, and measured context windows to help small models perform reliably. Plugins extend functionality with their own interfaces and runtimes, and can be developed and loaded from source within Shun. The system emphasizes durable context, real tools, parallel tasks, explicit permissions, and background process management, aiming to support long sessions of real work without degradation.
Shun is distributed as signed installers for macOS, Windows, and Linux, and it does not install additional developer tools, relying on the user's existing shell environment.
Pricing
Free
Detailed plans have not been confirmed in our catalog. Check the official website for current limits and billing terms.
Visit WebsitePrices and limits may change. Confirm the currency, billing period, seat minimum and usage caps on the official website.
Use Cases
- Developers use Shun to run coding agents locally on their own GPU without cloud dependency.
- Teams use Shun to keep code and transcripts private by construction.
- Users use Shun to manage context efficiently for small models with bounded tool outputs.
- Developers use Shun to create and load plugins from source without leaving the environment.
- Teams use Shun to run parallel tasks with isolated sessions and cancellation.
- Users use Shun to supervise background processes like servers and watchers.
- Developers use Shun to work with real filesystem and shell tools without sandboxed mocks.
Features
Run coding agents on local hardware
Keep code and transcripts local
Manage context with bounded tool outputs
Extend the harness with source plugins
Run parallel tasks in isolated sessions
Supervise background servers and watchers
Control agent actions with explicit permissions
Reviews
No reviews yet. Be the first to share your experience!