claake code
§ 01 — what it is, what it does

A desktop AI IDE for developers who want control.

Claake Code combines an editor, terminal, file tree, chat and AI coding agent in one native workspace — with every tool, skill, MCP server and provider visible and editable.

What is it?

An open-source AI coding IDE that exposes the full agent harness instead of hiding it behind a prompt box: tools, skills, MCP, providers, database sources and workspace context all stay visible.

Who is it for?

Developers and technical teams who want a powerful AI coding agent while keeping permissions, providers, tools and context fully auditable.

What makes it different?

Unlike black-box AI coding tools, every part of the operating surface can be inspected, edited and restricted before the agent touches your code.

§ 02 — core features

Core features for an auditable AI coding agent.

Claake Code is built around explicit controls, visible permissions and portable workflows rather than hidden prompts or opaque automation.

IDE
desktop AI IDE

Native workspace

Claake Code combines editor, terminal, file tree, chat and settings in a native desktop app for macOS, Windows and Linux.

AI
visible agent

AI coding agent

The agent can plan, edit files, run commands, inspect outputs, ask questions and coordinate work while keeping every action visible.

◴
editable tools

Tools you control

Read, glob, grep, edit_file, write_file, bash, web, question, todo, database, MCP, skills, sub-agents and teams can be toggled or described differently.

❀
skills

Markdown playbooks

Skills are local Markdown instructions in ~/.claakecode/skills that save repeatable expertise without a plugin store.

MCP
MCP support

MCP servers

Workspace-scoped Model Context Protocol servers attach external capabilities while remaining part of the visible agent surface.

⇄
providers

Multi-provider models

Connect Anthropic, OpenAI, Google, Kimi, OpenRouter and Mistral-style providers, then choose models for chats, sub-agents or teams.

DB
database-aware

Database safeguards

Approved Postgres, MySQL, SQLite, MSSQL or Supabase REST sources can be listed, described and queried with row limits and confirmations.

↺
rollback

Rollback and cleanup

Rollback checkpoints, context cleaning, conversation compaction and todo boards help long AI coding sessions stay reversible and readable.

§ 03 — product proof

A real IDE with an agentic core.

Open a repo, pick a provider, decide which tools are allowed, then let the agent plan, edit, run, inspect databases and coordinate sub-agents — with every move visible.

  • ↳Native desktop workspace. Monaco editor, xterm terminal, file tree, chat and settings live together.
  • ↳Provider freedom. Use OAuth subscriptions where available or API keys/OpenRouter when you want fully sanctioned routing.
  • ↳Workspace memory. AGENTS.md and DESIGN.md are injected as project instructions, so the assistant respects how you build.
  • ↳Safety rails. Database row limits, read-only flags, destructive-action confirmation and rollback checkpoints keep you in control.
Palette