Introduction
Harnez is a small harness for coding agents. It runs the agent loop in a server process and provides a terminal UI for interacting with it.
Harnez is model agnostic. You can connect any provider supported by the underlying agent core. It does not include a dashboard, plugin marketplace, or configuration that you did not request.
bun packages/server/src/index.tsWhy a harness instead of a framework
Harnez focuses on terminal-first coding agents. It is not a general extension platform. It has no plugin marketplace or public SDK. The code is modular internally, but the project does not provide a public extension API.
Where to go next
Start with Installation to get the binary running, then read Architecture to see how the pieces fit together.
Design principles
- Coding first, not coding only. Software development has the strongest user experience and tooling, but the runtime does not require it. You can also use Harnez for research, document analysis, and other agent workflows.
- Model agnostic. Models are runtime dependencies, not architectural foundations. Agents in the same session can use different models.
- Trusted execution. Harnez does not require approval prompts for ordinary reads, writes, edits, or shell commands. You control the session through visibility, steering, and interruption.
- Context is scarce, but execution state is not. Harnez stores large tool output outside the model's context window.