Documentation

Install Gradez and run your first optimization loop.

gradez discovers a measurable target, creates a baseline, runs parallel experiments in isolated git worktrees, and keeps only patches that improve your metric and pass your gates.

Quickstart

Run these commands in any Python project with git initialized.

pip Install Gradez gradez install cursor gradez discover "make the parser faster" gradez optimize gradez dashboard

After gradez install, Gradez offers a one-time interactive setup for provider, model, and API key (or skip and use flags). Run gradez configure anytime to update saved settings.

CLI reference

CommandDescriptionCommon flags
gradez installInstall Gradez rules or skills into a coding host.cursor, claude-code, --no-setup
gradez configureSet up or update LLM provider, model, and API key.none
gradez modelsList recommended models for a provider.--provider
gradez doctorCheck whether your environment is ready.cursor, claude-code
gradez dashboardStart the live dashboard on localhost.--port, --no-browser
gradez upgradeCheck PyPI for a newer Gradez release.none
gradez discoverAnalyze the repo and capture a baseline score (Python or Node).--provider, --model, --api-key
gradez optimizeRun the parallel optimization loop.--subagents, --budget, --stall
gradez statsShow analytics, token usage, cost estimate; export JSON or CSV.--json, --csv, --run-id
gradez tracesView experiment trace timeline.--experiment, --json
gradez reportExport a shareable HTML run report.--output, --run-id
gradez historyList recent optimization runs.--limit

Provider setup

Use CLI flags for a quick start, run gradez configure for interactive setup, or store defaults in ~/.gradez/config.toml.

[llm] provider = "groq" model = "llama-3.3-70b-versatile" api_key = "gsk_..." # Other providers: anthropic, openai, ollama, openrouter

OpenRouter

Try Gradez with free-tier models using one API key. Best for testing discover.

Ollama

Fully local. Run ollama serve and pull a model before starting.

Anthropic

Use Claude models when you need deeper code reasoning.

OpenAI

Use GPT models through the OpenAI-compatible path.

Dashboard

The dashboard gives you trust while the optimizer runs: latest run, experiments, pass rate, score chart, and trace logs.

gradez dashboard # Open http://127.0.0.1:8080
  • Experiment table shows status, score, delta from baseline, and gate result.
  • Score chart plots only passed experiments.
  • Analytics cards show pass rate, improvement, best score, and duration.

Analytics

Use stats and history to inspect run quality, export data, and prove that changes improved real metrics.

gradez stats gradez stats --json run.json --csv run.csv gradez report -o run-report.html gradez history --limit 20

Host plugins

Gradez works directly from the CLI, but it can also install guidance into coding hosts.

HostInstall commandWhat it adds
Cursorgradez install cursorProject rules in .cursor/rules/.
Claude Codegradez install claude-codeDiscover and optimize skill files.