Claude Code Haha is a desktop workspace for Claude Code on macOS, Windows, and Linux. It gathers multi-session tabs, project switching, a terminal entry point, session history, and search across every session into one application, putting a graphical layer over work that otherwise happens in a terminal.
A session can start on a chosen repository branch, in either the current working tree or an isolated worktree. Each turn’s changes are listed file by file with syntax-highlighted diffs and can be undone as a whole, and a built-in browser preview renders the page the agent just edited, cookies and login state included. Five permission modes run from asking every time to skipping permissions, with risky commands, tool calls, and follow-up questions approved in the interface.
Model choice is open: sign in to Claude, ChatGPT, or Grok, use presets for DeepSeek, Kimi, and Zhipu GLM, or point the app at LM Studio or Ollama running locally. Every model request is logged locally with status and timing, searchable and filterable when a call stalls or fails. Image generation runs in the chat through a signed-in provider or any OpenAI-compatible images API.
Multi-agent work has dedicated surfaces: a graphical subagent manager covering model, tools, and permission mode; an Agent Teams workbench showing members, tasks, a communication feed, and a dependency-lane canvas; and dynamic workflow orchestration, where the model writes and runs orchestration scripts that drive subagents concurrently or in pipelines, with phase views, interrupts, and resume.
Sessions do not have to stay on the desktop. Scanning a QR code continues a session in a phone browser, and locking the screen does not kill a running task, while chat integrations for Telegram, Feishu, WeChat, DingTalk, and WhatsApp allow chatting, switching projects, and approving actions from a messaging app. The app is built in TypeScript on Electron with a React and Vite interface, a Bun local runtime, an Ink terminal interface, and the Anthropic SDK, and speaks both MCP and the Language Server Protocol.
Features
- Multi-session workspace: tabs, project switching, a terminal entry point, session history, and a resizable sidebar
- Global search: one shortcut searches across every session and jumps to the match
- Branch and worktree launch: pick a repository branch and run in the working tree or an isolated worktree
- Diff review: per-file syntax-highlighted diffs of the current turn, with whole-turn undo
- Built-in browser preview: the page the agent just edited renders inside the app with cookies and login state
- Five permission modes: from ask-every-time to skip-permissions, with approvals handled in the interface
- Bring your own model: Claude, ChatGPT, or Grok sign-in, presets for DeepSeek, Kimi, and GLM, or local LM Studio and Ollama endpoints
- Visual MCP manager: add and edit STDIO, Streamable HTTP, and SSE servers at project, shared, or global scope
- Subagents and Agent Teams: a graphical subagent editor plus a workbench with members, tasks, a communication feed, and a dependency canvas
- Computer Use: screenshots, clicks, typing, and desktop application control after authorization
- Skill marketplace: discover, preview, and install third-party skills with source and safety status shown up front
- Remote access: continue a session in a phone browser by QR code, or through Telegram, Feishu, WeChat, DingTalk, and WhatsApp
