Skip to content

Getting Started

  • Open Agent Kit installed (pipx install oak-ci)
  • Ollama (Recommended for local embeddings) OR any OpenAI-compatible embedding service

Run oak init in your project directory:

Terminal window
oak init

The interactive setup walks you through two screens:

  1. Agent selection — Choose which AI coding agents to configure (Claude, Cursor, Gemini, etc.)
  2. Language selection — Choose which programming languages to install AST parsers for

What this does automatically:

  • Installs the CI daemon and CLI tools
  • Installs AST parsers (tree-sitter) for the selected languages
  • Configures Agent Hooks for supported agents
  • Registers MCP Servers for agents that support it
  • Updates IDE Settings (VSCode/Cursor) for optimal integration

The CI feature runs as a background daemon. Start it and open the dashboard:

Terminal window
oak ci start --open

On the first run, the daemon will:

  1. Scan your codebase to detect source files
  2. Build the index — Begin indexing your codebase using the installed parsers (may take a few minutes for large projects)

The dashboard opens automatically in your browser — you can watch indexing progress in real time.

After the daemon starts, use the dashboard to configure everything. Open the Configuration page from the sidebar to:

  • Choose your embedding provider — Select from Ollama, LM Studio, or any OpenAI-compatible endpoint. The UI auto-detects available models and dimensions.
  • Enable summarization — Optionally connect a local LLM for automatic session summaries.
  • Tune session quality — Control when background jobs process sessions:
    • min_activities: minimum activity count before a session qualifies for background processing (prevents summarizing empty or trivial sessions)
    • stale_timeout: how long an inactive session sits before cleanup by background maintenance
  • Manage exclusions — Add directory patterns to skip during indexing (e.g., dist/**, vendor/**). OAK includes sensible defaults (.git, node_modules, __pycache__, etc.) and respects .gitignore.

The Configuration page with embedding and summarization settings

Once OAK and your agents are set up, use the /project-governance skill to create a project constitution — a living document that codifies your team’s engineering standards so AI agents follow them consistently.

From any configured agent, run:

/project-governance Let's establish our project's constitution

The agent will:

  1. Analyze your codebase for existing patterns (testing, linting, CI/CD, project structure)
  2. Check for existing agent instruction files (CLAUDE.md, AGENTS.md, .cursorrules, etc.) and use them as context
  3. Create oak/constitution.md with standards tailored to your project
  4. Update agent instruction files to reference the constitution (additively — your existing rules are preserved)

This works in both greenfield projects (no existing rules) and brownfield projects (existing conventions that should be formalized).

As your project evolves, use the same skill to amend the constitution when new rules are discovered:

/project-governance Let's ensure we always follow the vertical slice architectural pattern

After updating the constitution, sync the changes to all agent instruction files:

Terminal window
oak rules sync-agents

See the Skills page for the full /project-governance reference.

Open the dashboard home page to verify everything is working:

  • Files Indexed: Shows the number of source files tracked
  • Memories: Count of stored observations
  • Sessions: Agent sessions being tracked
  • System Health: Embedding and summarization provider status

You can also check from the terminal:

Terminal window
oak ci status