How Dash Works
You give Dash a task, a repo, and a test command. It does the rest:
- Understand — reads your codebase, searches for relevant files, maps conventions
- Plan — writes a test spec and implementation strategy before touching any code
- Implement — generates search/replace diffs applied cleanly via git
- Verify — runs your test suite. If tests fail, diagnoses the output and retries — up to 3 times
Access & Pricing
Dash orchestrates your inference. Choose the approach that fits your workflow:
BYOK
Free
Connect your own OpenRouter API key. Pay model costs directly — typically fractions of a dollar per task. No markup, no subscription, no rate limits.
Local
Free
Point Dash at LM Studio running on your localhost. No API key, no account required.
Hosted
Coming soon
Dash provides the inference with optimized model defaults. No key management, no model selection — just run tasks.
Calling Dash from your agent
Dash is designed to be invoked by AI orchestrators — hand off implementation tasks and stay focused on the bigger picture. Run the built-in guide to get integration instructions:
npx @jkershaw/dash@latest guide
Tip: Always include @latest to ensure you have the current version. Without it, npx may serve a cached older copy.
- Dash works best on 1–4 file, single-concern tasks. Decompose larger work first.
- Use
-y (auto-approve) and -v (verbose) flags when calling from an agent.
- Describe the goal, not the steps. Dash discovers the codebase and creates its own plan. Over-specified prompts cause complexity rejections.
- For full CLI reference:
npx @jkershaw/dash@latest help
