OpenClaude vs OpenCockpit: Claude Code on Any LLM
Both bring a Claude Code-style agent to any model. OpenClaude is a terminal CLI; OpenCockpit is a browser workbench. How they differ and when to use which.
If you've searched for an "OpenClaude GUI", you probably already know what you want: the Claude Code way of working, on a model that isn't Claude. OpenClaude and OpenCockpit both answer that. They take very different routes to get there, and the right pick depends on where you like to work.
Facts about OpenClaude below come from its README as of October 2026. If something has changed, the README wins.
OpenClaude: the CLI, opened up
OpenClaude is an open-source coding-agent CLI — "runs anywhere, uses anything". Its README says it originated from the Claude Code codebase and has since been substantially modified to support many providers: OpenAI-compatible APIs, Gemini, GitHub Models, Codex, Ollama and a long list of others.
Because it keeps the Claude Code shape, it keeps the Claude Code feature set: MCP servers, subagents, slash commands, a terminal-first workflow, plus a bundled VS Code extension. You install it with npm install -g @gitlawb/openclaude and use it where you'd use claude.
On licensing, the README is specific: MIT for OpenClaude contributors' modifications, while the Claude Code code it derives from remains Anthropic's.
OpenCockpit: a workbench around the agent
OpenCockpit doesn't fork any CLI. It's a local web app — Agent, Explorer and Console panels in one browser window — and each Agent tab picks an engine:
- Claude tabs run on the official Claude Agent SDK, with your existing
claudelogin. - Codex tabs drive OpenAI's
codexCLI. - DeepSeek, GLM, Kimi and Ollama tabs run OpenCockpit's own Built-in Agent against the provider's API.
The point isn't one agent on many models so much as many sessions side by side: several projects in the sidebar, several tabs per project, each on whatever engine suits it, with the file browser, diffs, terminals and per-tool-call snapshots right next to the chat. It's also client–server, so you can run it once on a shared dev box and give every teammate a seat. MIT licensed.
Side by side
| OpenClaude | OpenCockpit | |
|---|---|---|
| Interface | Terminal CLI + VS Code extension | Browser workbench (Agent / Explorer / Console) |
| How it reaches other models | Its own CLI, many providers | Per-tab engine: Claude, Codex, DeepSeek, GLM, Kimi, Ollama |
| Provider breadth | Wider — 20+ providers incl. Gemini, GitHub Models, LM Studio | Six engines |
| MCP & subagents on non-Anthropic models | Yes | No — the Built-in Agent has seven tools |
| Parallel work | One session per terminal | Many projects × many tabs in one window |
| Reviewing what the agent changed | Your usual git workflow | Per-tool-call snapshots and a diff viewer built in |
| Team use | Each person runs their own CLI | Self-host once, every teammate gets a seat |
| License | MIT for its modifications; derived code remains Anthropic's | MIT |
Which one, then?
Pick OpenClaude if you live in the terminal, want the broadest choice of providers, or need MCP servers and subagents on a non-Anthropic model. On those points it simply does more.
Pick OpenCockpit if what you're missing is the GUI part: running several projects and models in parallel, seeing files and diffs next to the conversation, reviewing every tool call, or sharing one setup with a team.
They don't exclude each other, either. OpenCockpit's Console panel is a real terminal, so you can run openclaude in one bubble while the Agent panel works on something else.
Try OpenCockpit
npm i -g @surething/cockpit && cockpit
Engine setup is in AI Engines, and the longer comparison with the official CLI and Desktop app is in Claude Code CLI vs GUI vs Desktop.