Hive

Hive is an AI coding agent orchestration desktop workstation designed for parallel coding scenarios. Its core mechanism is a Worktree-First workflow—each AI agent runs in an isolated Git Worktree branch, completely eliminating file conflicts and context interference between agents.

Core Capabilities#

Worktree Isolation & Parallelism

• Each agent binds to an independent Git Worktree branch—no stash or branch switching needed
• One-click create/archive/delete Worktrees, auto-assigned city names for easy identification
• Worktree Connections: bidirectional cross-branch linking with real-time file change sync, quick switching, diff comparison, shared terminals, and AI context sharing

AI Agent Orchestration

• Supports Claude Code, OpenCode, Codex, and local LLMs as coding agents
• Real-time streaming responses, tool call visualization, permission approval, undo/redo
• Kanban board: task cards can be directly assigned to AI agents, tracking Todo → In Progress → Done
• Cross-repository AI context: link frontend/backend/shared type repos for full-stack awareness in a single session

Engineering Tool Integration

• Smart file browser: real-time Git status indicators, inline diff, file history browsing
• Monaco Editor kernel for VS Code-level editing experience
• Full Git visual operations: commit/push/pull/branch management, staged files and commit history viewing
• MCP protocol extension via .mcp.json, Claude Code command templates via .claude/commands/

Experience & Efficiency

• Command palette (Cmd+K): search sessions, switch Worktrees, execute commands
• Spaces: organize related projects and Worktrees into logical workspaces with pinning
• 10 built-in themes (6 dark, 4 light), system theme following supported

Use Cases#

• Multi-feature/multi-bug parallel development with isolated Worktrees per agent
• Full-stack development: connect frontend, backend, and shared type repos
• Code review: connect reviewer and author Worktrees for real-time diff comparison
• Refactoring verification: connect old and new implementations to ensure feature parity

Installation & Quick Start#

macOS (Homebrew):

brew install --cask hive-app

Cross-platform download: Get platform-specific installers (.dmg / .exe / .AppImage / .deb) from GitHub Releases.

Three steps to start:

1. Open Hive → Click "Add Project" → Select a local Git repository
2. Select project → Click "New Worktree" → Choose or create a branch
3. Open Worktree → Click "New Session" → Select an agent to start coding

Architecture Overview#

Built on Electron + electron-vite, TypeScript at 97.3%, using pnpm for package management, Vitest for testing, ESLint + Prettier for code style, and mise for task running. Project structure includes src/ (source), test/ (tests), docs/ (documentation), scripts/ (scripts), resources/ (resources), etc.

Unconfirmed Information#

• Initial release date: README only states MIT © 2024, exact launch date not specified
• Full list of supported LLM providers: website mentions local LLMs but doesn't list specific compatible providers
• Plugin system and cloud sync/team collaboration: both listed as future features on the roadmap, not currently available