Providers — Claude, Codex, Gemini, and more, side by side
Claude, Codex, Gemini, OpenRouter, local models — every pane with its own provider.
Any CLI, any model, any pane
A provider defines how each pane connects to an LLM. Overclock ships Claude OAuth, Codex, and Gemini ready to go — detected on your machine, using the login that's already there — and opens the door to the rest: OpenRouter, Fireworks, MIMO, DeepSeek, GLM, Kimi, and local models via Ollama, LM Studio, or llama.cpp.
The point isn't the list — it's the granularity: each pane can use a different provider and model, with its own permission mode. A mixed team of Claude, Codex, and a local model on the same grid is the normal case, not the trick.
How to use it
Open Settings → Providers
the detected CLIs already show up; add the API-key ones with a connectivity test on setup.
Choose it in + PANE
provider and model per pane; the header shows the routing (e.g., CLAUDE ← MIMO).
Mix them on the same grid
every agent with its own provider, its own model, and its own permission mode, side by side.
What connects today
| Provider | How it connects | Note |
|---|---|---|
| Claude | OAuth from the local session | Multi-account with isolation and a per-account proxy |
| Codex | CLI detected on the PATH | Models listed straight from the installed CLI |
| Gemini | CLI detected on the PATH | — |
| Antigravity | CLI with its own isolation | — |
| OpenRouter · Fireworks · MIMO | API key | Hundreds of models; region-based routing on MIMO |
| DeepSeek · GLM · Moonshot/Kimi | API key | Dedicated templates with two routes |
| Local | Ollama, LM Studio, llama.cpp | Models running on your hardware |
What this area has today
Claude OAuth built-inshipped
Uses the Claude Code session that's already on your machine — no manual API key.
Codex + Gemini built-inshipped
Detected on the PATH, with models listed straight from the installed CLI — not a static list.
Antigravityshipped
Google's CLI as a pane provider, with its own isolation.
OpenRouter · Fireworks · MIMOshipped
Hundreds of models via API key — with a connectivity test on setup.
Local modelsshipped
Ollama, LM Studio, and llama.cpp — your hardware, your models, zero cloud.
Permission mode per paneshipped
Danger, Auto, Pend, or Plan — the autonomy level is chosen pane by pane.
Multi-account Claude + proxyshipped
Several isolated Claude accounts, each able to exit through its own IP via a per-account proxy.
Kimi · Cursor · Copilotpartial
Support is present in pane spawn, not yet validated in real use — to be honest, that's what we have today.
Status straight from the product's brain: shipped = actually in use; partial = exists, with documented limits.
Where this actually gets to work
all squads →Showcase squads arrive as they're recorded — see what's already published at /squads.