Stave

Stave is a desktop AI coding workspace for Claude, Codex, Cursor Agent, Kiro CLI, and OpenCode. It combines task-oriented chat, repo-aware context, an editor and terminal, workspace memory, and local automation in one app.

Install the App on macOS

The packaged macOS install flow uses GitHub CLI authentication:

gh api -H 'Accept: application/vnd.github.v3.raw+json' repos/sendbird/stave/contents/scripts/install-latest-release.sh | bash

If this is your first time using gh, or you need SSO or scope help, see the full Install Guide.

Packaged macOS builds can also check for updates from the top bar and install them in place.

Build from Source

Requirements:

• Bun
• Node.js >= 20
• a native build toolchain for better-sqlite3 and node-pty

Install dependencies:

bun install

Start the Electron app in development:

bun run dev:desktop

Useful command for validating a packaged desktop build:

bun run run:desktop:packaged:logged

This rebuilds Electron native dependencies, produces the desktop build, launches the packaged app, and writes a timestamped log file for debugging packaged-only issues.

For other development, packaging, and contribution commands, see Developer and Contributing Guide.

Provider Setup

To use provider-backed chats, install and authenticate the CLIs you want Stave to drive.

claude auth login
codex login
agent login
kiro-cli login
opencode auth login

Recommended next steps:

• Open Settings -> Providers and choose the runtime controls you want.
• Open Settings -> MCP if you want to enable the built-in local MCP server.
• If macOS asks for Desktop, Documents, or Downloads access, approve it once or see macOS Folder Access Prompts.

Features

• English (default) and Korean throughout the app; switch in Settings or the home menu, with the choice saved across restarts
• task-based Claude, Codex, Cursor Agent, Kiro CLI, and OpenCode chats with approvals, diffs, and queued follow-ups
• usage-limit pauses with manual or reset-time resume, and restored queues that wait for you after a restart
• other tasks attached as context through title search or a sidebar drag, with a choice of the latest reply or recent conversation
• Monaco editor, docked terminal, quick open, command palette, and source control actions
• Lens browser panel for inspecting a live page and pulling DOM, console, or element context into a task draft
• Compare Runs with candidate and judge turns, and Review tasks that run a read-only review on the model you pick beside the task and hand back only the findings, with configurable review presets, installed skills or custom prompts
• Crane connector for queuing repository issues into approval-gated local Claude or Codex runs
• Issues surface listing assigned Crane and Jira Cloud tickets with one-click local kickoff
• workspace documents (plans, reports, specs) that agents revise in place, with every revision recorded
• workspace-scoped notes, todos, PR links, Jira, Figma, Confluence, and Slack references
• scheduled Claude and Codex automations with per-run results, repository selection, and reusable Information context
• hand a goal to an agent and leave it: an Auto agent's run keeps going through approvals, turn limits, unanswered questions, pull request checks and usage resets, finishes only when the workspace check Stave runs itself passes, and calls you only for a decision (how to work this way)
• agents with a workflow: the stages you would otherwise prompt one by one — Stave opens the draft PR, watches checks, and checks in with you only where the agent says
• Agent performance for comparing delegated runs across workspaces, and task Outputs for inspecting saved answers and changes
• git worktree-aware repository and workspace management
• editable workspace kickoff proposals from external sources and prompts
• Fleet Action required inbox for questions, approvals, agent run sign-offs, failed runs, results, and PR blockers across every workspace
• notifications, attachments, skill selection, custom model shortcuts, and theme presets
• local-only MCP access for same-machine automation and tool-driven workflows

Documentation

• Install Guide for the full macOS install and update flow
• Lens Browser Guide for inspecting a live page and sending its context into a task draft
• Provider Browser Access for using @web with the active provider's native browser extension
• Provider Sandbox and Approval Guide for runtime safety and second opinions
• Workspace Documents for plans and reports an agent writes and revises
• Turn Activity for queued follow-ups, restart recovery, and resuming work after a usage limit
• Attachments for files, images, and other tasks used as context
• Review Tasks for reviews and second opinions that run in their own read-only task
• Agent Performance and Task Outputs for choosing between cross-workspace metrics and a task's saved outputs
• Local MCP User Guide for same-machine automation setup
• Crane Connector Guide for pairing Crane with this Stave installation and approving issue runs locally
• Issues Guide for reviewing assigned Crane and Jira tickets and kicking one off locally

For Developers And Contributors

Development setup, build and packaging commands, architecture pointers, and contribution guidance live in Developer Docs and the Developer and Contributing Guide.

License

Apache-2.0 — see NOTICE for attribution.