Loading…
Loading…
Provider catalog
A cloud provider needs a key you own; a local runtime needs a URL for a server you already run. The label, the default model, the model count and the list price on every card are read from the shared model catalog and the key list the apps ship with, so this page cannot name a provider they do not.
The roster
These 14 providers accept a key you hold, bill you on your own account, and between them carry 36 catalogued models. The price is the provider’s base list rate per million tokens as the catalog records it; individual models and long-input bands can sit above or below it.
Desktop also talks to Ollama, LM Studio, llama.cpp, and vLLM. None of them carry catalogued models, because AGI asks the server you started what it is holding rather than assuming.
The source file
packages/contracts/types/src/models.json is embedded in the CLI and in the Desktop Rust runtime with include_str!, and the web app imports the same module. Adding a provider or moving a price moves every surface and this page at once. Below is what the CLI prints from that file; the CLI surface itself is coming soon.
$ agi models list OPENROUTER: openrouter-auto [TVR] 2M ctx — out $0.00/$0.00 base openrouter-free [TVR] 200K ctx — out $0.00/$0.00 baseFlags: T=tools, V=vision, R=reasoning. !=deprecated, B=beta.
Inside one session
/model resolves the provider from the catalog, then compares the route that model needs against the privacy authority already written beside the transcript. Anthropic to Google is a swap and the thread carries on. Anything that would change the authority is refused by name, and the reviewed fork that would do it properly is described on the Local page.
Providers
Desktop holds the encrypted key store and every local runtime URL (Linux assets · v1.2.0). The CLI compiles the same catalog and is coming soon.
Public launch: date to be announced