Kaitoi CLI.
Install Kaitoi CLI, work with Code and Crew, connect editors and local services, and automate sessions or recipes with practical guides and API references.
Getting started
- Install and update Install Kaitoi CLI on macOS, Windows, or Linux, choose the command for your work, and learn how to check, download, and apply updates.
- Your first Kaitoi Code session Start Kaitoi Code in a project, inspect it in Plan mode, review a small edit in Build mode, and check the result before moving on.
- Sign-in and configuration Sign in to Kaitoi CLI, save your Code preferences, find the right settings file, and understand which choices apply to a new session.
- Files and state on disk Find Code settings, session logs, audit records, agent worktrees and Crew evidence, and configure project file exclusions.
- Hosted and local models Choose a hosted model or connect your own model server in Kaitoi Code. Learn model selection, local setup, context limits, and where requests go.
- Local model setup Set up Ollama, LM Studio or an OpenAI-compatible model server for Kaitoi Code, with no Kaitoi login or Kaitoi inference charge for local-model turns.
- Costs and budgets Read Code usage, set a credit or step limit, and distinguish model spending, tool charges, external engines, and the limits of a running budget.
Working with Kaitoi Code
- Plan, Build and Crew modes Choose Plan for inspection, Build for approved changes, or Crew for durable delivery with its own reviewed authority and bounds on delegated work.
- Permissions and the sandbox How Kaitoi Code asks before it acts. Trusting a folder, approval cards with diffs, rules you save per project, auto-approve, and the sandbox around commands.
- Files, attachments and edits Give Code the right files and images, use line-specific mentions, review edits, and understand the boundary between workspace access and attachments.
- Commands and background processes Enable model command tools, inspect sandbox and approval details, use your own shell command, and track or stop the background processes Code owns.
- Sessions, history and context Resume or fork a conversation, inspect context pressure, compact long sessions, export useful records, and distinguish a fresh session from file recovery.
- Checkpoints, undo and rewind Recover from an unwanted Code exchange, choose whether files or conversation should go back, and understand skipped files, redo, and checkpoint limits.
- Web research and media generation Research with cited public sources, prepare a media brief, review a generation quote, and recover an existing result when downloading or saving fails.
- Kaitoi projects, graphs and custom nodes Build Kaitoi workflows from Code, inspect real node schemas, distinguish inline runs from saved projects, and validate custom nodes before installation.
- Keyboard shortcuts and slash commands Find Code's keyboard shortcuts and command families, distinguish steering from queueing, and understand how Escape behaves at a prompt or approval card.
Extending Kaitoi Code
- Extensions overview Extend Kaitoi Code with MCP servers, skills in SKILL.md folders, your own slash commands and hooks, each under your control.
- MCP servers Add external tools to Code, resolve user and project server settings, approve the launch, sign in when required, and set separate per-tool grants.
- Skills and project commands Package reusable instructions as a skill, add a project slash command, understand discovery precedence, and keep prompt expansion separate from execution.
- Hooks Run reviewed automation at Code events, configure hook matching and timeouts, and distinguish blocking decisions from context and after-the-fact feedback.
- Project instructions and memory Put durable project guidance where Code can load it, understand parent and nested instruction scope, and keep bounded project memory concise and current.
Subagents and Crew
- Subagents and background agents Delegate a bounded task, choose a role and lifetime, monitor background agents, and distinguish shared-workspace work from worktree isolation and Crew.
- Claude Code and Codex engines Assign subagent roles to Kaitoi, local, Claude Code, or Codex engines, understand credential and billing boundaries, and review the engine's start grant.
- Your first Crew run Define a bounded coding task for Crew, choose its engines and verification commands, and review workspace preparation, checkout mirroring, and budgets.
- Monitor, steer and resume runs Follow the correct Crew run, distinguish pause, interrupt, cancel, and continuation, and use fresh state versions when controlling a durable run.
- Worktrees, verification and results Inspect a Crew candidate and its evidence, understand guarded checkout mirroring, and keep saved work, verified completion, integration, and release distinct.
- Crew command reference Look up Crew commands, run flags, spending limits, continuation grants and recorded phases before starting or recovering a coding run.
Editors and automation
- VS Code extension Connect the VS Code chat view to your installed Kaitoi CLI, add selected code as context, and handle workspace trust, permissions, diffs, and cancellation.
- Connect other editors with ACP Configure an ACP editor to launch Kaitoi Code over standard input and output, resume saved sessions, and handle authentication, context, and permissions.
- Headless runs and JSON events Run one Code turn from a prompt or standard input, constrain workspace authority and cost, and consume JSON events without confusing them with session SSE.
- Local session HTTP API Start a loopback Code server, create trusted sessions, submit messages, follow resumable events, and answer permissions using the local session API.
- Python SDK Connect Python to a running local Code server, iterate one turn or follow a session, answer specific permissions, and cancel work with the correct lifecycle.
- GitHub Actions Configure a bounded GitHub integration, understand comment triggers and runner authority, and distinguish a dry run from remote comments, pushes, and PRs.
Local services and recipes
- Connect local services Bring a local Runtime or model service into Kaitoi through a tunnel, distinguish service health from readiness, and stop the resources you actually started.
- Install and run recipes Find recipes, inspect requirements, install and run them with declared inputs, and connect their outputs to Studio while keeping local and cloud files distinct.
- Create recipes Define a recipe with YAML and a companion script, map inputs to command arguments, collect declared outputs, and distinguish generation from verified execution.
- Runtime API Submit local recipe jobs, poll their status, upload inputs and retrieve outputs, and distinguish Runtime's short-lived job store from Code's session API.