The open Claude Code GUI · Bring any agent
One seat. One AI. Everything under control.
The open-source Claude Code GUI, built like an IDE for the whole dev loop — code, terminal, browser & DB in one workbench. Runs on your laptop, or on a shared dev box where every teammate gets a seat.
Watch the 24-second tour
Why Cockpit beats raw Claude Code
Run 5+ agent sessions across projects at once. Get notified when each finishes — no terminal juggling.
`claude` CLI works out of the box. For Codex, DeepSeek, Kimi or Ollama, paste a key (or none for Ollama). Everything runs locally.
A real xterm.js terminal, Chrome automation, PostgreSQL / MySQL / Redis — all inside one window your agent can drive.
AI chat that scales with you
- Claude out of the box via the official Agent SDK; Codex / DeepSeek / Kimi / Ollama via per-tab BYOK
- Multi-project concurrent sessions with desktop notifications
- Session pinning, forking, cross-project session browser
- !command runs shell from chat, output piped back as context
Code & files, all-in-one
- Directory / Recent / Git Changes / Git History in one tree
- Shiki syntax highlighting with Vi-mode editing
- Git blame, diff, branch switching, worktree
- LSP go-to-definition + Code Map call graph (TS/JS/Python/Go/Rust)
Terminal & smart Bubbles
- Full terminal emulator (xterm.js) with shell integration
- Browser Bubble — control Chrome via accessibility tree
- Database Bubbles — PostgreSQL, MySQL, Redis
- Drag-to-reorder bubbles, grid / maximized layout
Smart Bubbles in Console
Floating panes that connect to anything — controlled by AI or by you.
Click, type, navigate, screenshot, network inspection.
Browse schema, run queries, export data.
Browse databases & tables, run queries.
Browse keys, inspect values, execute commands.
Claude by default — bring any agent you want
Each engine runs in its own tab, with its own session history. Pick from the new-tab dropdown.
Full Claude Agent SDK — tools, plans, slash modes. Zero setup if `claude` is already configured.
Reuses your `~/.codex` config — same chat, shell and bubbles, different tab.
Anthropic-compatible endpoint via the Claude SDK. Paste a key, pick `v4-pro` or `v4-flash`.
Tool calls render in chat like Claude's — see calls and results inline.
Auto-starts the daemon; pick any pulled model. No key, no internet.
Everything runs locally.
Read the codebase like a map, not a tree
Get oriented in any unfamiliar codebase. Every function shows its body in the middle, who calls it on the left, what it calls on the right — click a pin to jump.
- New to a repo? See every function and its connections at a glance.
- Click a caller / callee pin to follow the trail — no more grep-and-pray.
- Trace a bug from entry point to root cause without losing your place.
- Works across TypeScript / JavaScript, Python, Go, Rust — no setup.
Open any file in Explorer → switch to Code Map view.
Give the AI a query graph, not a grep
The missing layer between your AI agent and your codebase. The same tree-sitter index behind Code Map, exposed as 10 HTTP endpoints — so the agent queries coordinates instead of grepping text and reading whole files.
- 10 endpoints, two tiers: 6 base shapes (where is X, who calls X, impact, co-edit…) + 4 analytics (context, related, risk, affected).
- Returns coordinates only (file / line / qname) — the agent Reads exact ranges, never whole files.
- Co-edit catches must-edit-together file pairs no static analysis sees — parallel registries, double-writes, sibling configs.
- Incrementally synced by a file watcher — no rebuild; current across mid-PR and bulk AI edits.
Type /cg in any chat to enter graph-exploration mode.
One slash, one AI mindset
Slash commands flip the agent into a specific posture — talk first, debug only, never touch code.
/qaClarifyRestate the requirement, ask back on anything ambiguous, follow KISS — talk first, never code.
/fxDiagnoseEvidence-chain bug analysis — reason through the failure end-to-end, never edit a file.
/exExploreStructured 6-step discussion: study → diverge → converge → verify → summarize. No mid-flow interruptions.
/goLandTake a converged plan, slice into MVP stages, code + self-verify each, recap at the end.
/cgCodeGraphGraph-first exploration — 10 endpoints answer symbol / callers / impact / risk questions. Precise where grep is fuzzy.
/ccCockpit CLIDrive the cockpit CLI — codegraph, terminal, browser subcommands, each self-documenting via --help.
Custom: drop any SKILL.md and add it via the Skills sidebar — it auto-appears in the autocomplete menu.
Stack commands into one workflow
Start several lines with / or @ and Cockpit reads the whole message as one ordered run — clarify, fix, then have a sub-agent review the fix, in a single send.
/runs in the main session@delegates the step to a sub-agent
/fxfind why retries double-charge the card@craudit the fix for race conditions
Ship faster, together
- LAN-shareable review pages — teammates need zero install
- Line-level comments with reply threads
- Send any comment back to AI as context for an automated fix
- Built on plain Markdown — no separate review system
Scheduled Tasks
One-time, interval, or cron-based scheduling. Pause, resume, reorder, track results across projects.
Skills
Drop in any SKILL.md to teach the agent a new trick — invoke with /skill-name from chat. Manage everything from a single Skills panel.
How OpenCockpit compares
An honest snapshot as of July 2026 — each tool wins somewhere. Spotted an error? PRs welcome.
| OpenCockpit | Claude Code Desktop (official) | Opcode | |
|---|---|---|---|
| Positioning | IDE-like workbench for the whole dev loop | Agent session companion | Session manager for Claude Code |
| Architecture | ✅ Web client–server — self-host on a shared dev box, a seat for every teammate | Single-user desktop app | Single-user desktop app |
| Open source | ✅ MIT | ❌ Closed source | ✅ AGPL-3.0 |
| Engines | ✅ Claude + Codex / DeepSeek / Kimi / Ollama (BYOK) | Claude only | Claude only |
| Parallel multi-project sessions | ✅ | ✅ | ✅ |
| Agent-drivable browser & DB | ✅ Smart Bubbles: Chrome / Postgres / MySQL / Redis | ❌ Preview pane only | ❌ |
| LAN-shared code review pages | ✅ | ❌ | ❌ |
| Fully offline / air-gapped | ✅ via Ollama | ❌ | ❌ |
| Phone / tablet access | ✅ Any LAN browser — code runs on your machine | ✅ Via cloud sandbox (Claude Code on the web) | ❌ Desktop only |
| Native desktop app | ❌ Local web app (needs Node ≥ 20) | ✅ | ✅ Tauri |
| Newest Claude Code features on day one | ⏳ Tracks Agent SDK releases with a lag | ✅ First party | ❌ |
| Automation triggers | One-time / interval / cron | ✅ Routines: cron + API + GitHub events | Background agents |
| Session checkpoints / rewind | Pinning & forking only | ✅ | ✅ Checkpoint timeline |
| Usage / cost analytics | Basic token counts | n/a (plan-based) | ✅ Full dashboard |
| Cost | GUI free (MIT); AI billed by whichever engine you bring — $0 with local Ollama | App free; needs a paid Claude plan or API billing | GUI free (AGPL); needs a paid Claude plan or API billing |
| Actively maintained | ✅ | ✅ | ⚠️ Last release Aug 2025 |
Pick Claude Code Desktop if you live inside the Anthropic ecosystem and want first-party polish.
Pick Opcode if you want a native desktop feel with checkpoints and cost analytics.
Pick OpenCockpit if you want more than a chat window — an open-source, IDE-like cockpit for the whole dev loop, with any engine you bring, even from your phone.
Built on the official Claude Agent SDK
Cockpit’s core uses Anthropic’s official Claude Agent SDK. If your `claude` CLI is configured, Cockpit works — no extra setup. Other engines ride the Vercel AI SDK and the same agent loop. All local.
Ready to fly?
Install once, then `cockpit` from any directory. (Or `cock` if you prefer the short alias.)