Any CLI, any model, any pane.
A provider defines how each pane connects to an LLM. Overclock ships with Claude OAuth, Codex, and Gemini ready to go — detected on your machine, with 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 CLIs you already use, now side by side
Each CLI below runs in its own pane, with the login that's already on your machine — each agent uses the subscriptions and tokens you already pay for, with no extra model cost beyond the subscriptions you already have. What's marked partial is exactly that: support is present, not yet validated in real use.
Claude Code
Local session OAuth — no manual API key.
Multi-account with isolation and a proxy per account. The cockpit's reference agent.
Codex
CLI detected on PATH.
Models listed straight from the installed CLI — not from a static list.
Gemini
CLI detected on PATH.
Google's CLI running as a pane, side by side with the others.
Antigravity
CLI as a pane provider.
With its own isolation per pane.
GrokPARTIAL
Support present in pane spawn.
Not yet validated in real use — honestly, that's where it stands today.
KimiPARTIAL
Support present in pane spawn.
Not yet validated in real use — honestly, that's where it stands today.
CursorPARTIAL
Support present in pane spawn.
Not yet validated in real use — honestly, that's where it stands today.
CopilotPARTIAL
Support present in pane spawn.
Not yet validated in real use — honestly, that's where it stands today.
Hundreds of models with one API key
Beyond the CLIs, Overclock connects model providers by API key — with a connectivity test at setup, before any pane depends on them. The pane header shows the routing (e.g., CLAUDE ← MIMO).
OpenRouter
API key, with a connectivity test at setup.
Hundreds of models behind a single endpoint.
Fireworks
API key, with a connectivity test at setup.
Open models served at production speed.
MIMO
API key, with a connectivity test at setup.
With routing by region.
DeepSeek
API key, dedicated template with two routes.
DeepSeek's models straight in the pane.
GLM
API key, dedicated template with two routes.
Z.ai's GLM family as a pane provider.
Moonshot / Kimi
API key, dedicated template with two routes.
Kimi's models via the Moonshot API.
Your hardware, your models, zero cloud
Ollama
Local server detected on your machine.
Your hardware, your models, zero cloud.
LM Studio
Local server detected on your machine.
Your hardware, your models, zero cloud.
llama.cpp
Local server detected on your machine.
Your hardware, your models, zero cloud.
Every pane with its own provider
The point isn't the list — it's the granularity: each pane can use a distinct provider and model, with its own permission mode (Danger, Auto, Pend, or Plan). A mixed team of Claude, Codex, and a local model on the same grid is the normal case, not the trick.
Open Settings → Providers
The detected CLIs already show up; add the API-key ones with a connectivity test at setup.
Choose it in + PANE
Provider and model per pane; the header shows the routing (e.g., CLAUDE ← MIMO).
Mix them in the same grid
Each agent with its own provider, its own model, and its own permission mode, side by side.
The complete anatomy of this area is at Features → Providers