Models & Providers
Claude Code and Codex, plus Anthropic, Bedrock, Z.AI, MiniMax, Kimi, custom endpoints and local Ollama.
Quack doesn't lock you into one backend.
Agentic CLIs
Quack drives a CLI you already have installed. Two backends:
- Claude Code —
claudeon your PATH. Full support: permission modes, plan mode, subagents. - Codex —
codexon your PATH. Models:gpt-5-codex(default),gpt-5,gpt-5.5,o3, plus a free-text field for anything else your OpenAI account can reach.
A chat is pinned to whichever CLI it started with for its whole life — switching means starting a new chat, since each CLI keeps its own session memory that doesn't carry over.
Providers for Claude Code
Claude Code speaks the Anthropic API, so anything that serves an Anthropic-compatible endpoint works. Quack ships these presets — you only paste an API key:
| Preset | Endpoint | Models | Context |
|---|---|---|---|
| Anthropic | official | Sonnet, Opus, Haiku | 200k |
| Z.AI | api.z.ai/api/anthropic | glm-4.6 / glm-4.5-air | 200k |
| MiniMax | api.minimax.io/anthropic | MiniMax-M2 | 1M |
| Kimi | api.moonshot.ai/anthropic | kimi-k2-0905-preview | 256k |
AWS Bedrock is also supported — pick a region and a model ID instead of a base URL.
Custom endpoints
Not on the list? Add your own. A custom provider takes a base URL, the model IDs to map onto Sonnet and Haiku, and a context window. You can also duplicate a built-in preset and edit the copy — handy for pointing at a proxy or a self-hosted gateway.
Local models
Ollama runs models on your own machine. No API key, no internet. Quack
looks for it at localhost:11434 and lists whatever you've pulled.
Add keys and endpoints in Settings → Providers.
Picking a model
The composer's model chip opens a quick picker; a fuller Model Browser shows every available model grouped by provider, with favorites and free-tier badges. Model names shown are stable aliases (Sonnet, Opus, Haiku) rather than raw version strings — so the label in your chat doesn't change every time a provider ships a point release.