models

Providers & models.

Squeezy keeps the code analysis, permissions, and accounting on your machine, and routes model turns through whichever provider you configure. You bring the key.

Choosing a route

Squeezy ships 27 provider ids. The quickest start is an aggregator: one API key reaches many hosted models. If you prefer a vendor's own API, a subscription sign-in, a cloud platform's catalog, an open-weight host, or a model running on your own machine, Squeezy supports those too. Set the provider and supply its credential; the tables below list the environment variable for each.

Aggregators, one key, many models

OpenRouter

OpenAI-compatible aggregator with many hosted models under one key. Pricing and cache support depend on the selected model.

OPENROUTER_API_KEY

Vercel AI Gateway

OpenAI-compatible gateway for hosted model access through Vercel.

AI_GATEWAY_API_KEY

PortKey

OpenAI-compatible gateway for virtual keys, routing, and observability.

PORTKEY_API_KEY

Cloudflare AI Gateway

Org-wide proxy with rate limiting and routing in front of your chosen models.

Cloudflare gateway

First-party vendor APIs

OpenAI

Native OpenAI route with usage parsing and cache-related request metadata where supported.

OPENAI_API_KEY

Anthropic

Native Anthropic route with API-key and OAuth paths plus cache read/write accounting where exposed.

ANTHROPIC_API_KEY

Google Gemini

Native Gemini route with API-key configuration and streaming usage metadata where available.

GEMINI_API_KEY

Subscription sign-in (OAuth)

OpenAI subscription

Use an OpenAI subscription through its OAuth sign-in instead of a metered API key.

OpenAI OAuth

GitHub Copilot

Use a GitHub Copilot subscription through its OAuth sign-in.

Copilot OAuth

Cloud-platform hosts

Amazon Bedrock

AWS-hosted models using the AWS credential chain and Bedrock runtime APIs.

AWS credentials

Azure OpenAI

Azure-hosted OpenAI with a deployment-specific endpoint and API-key or bearer-token configuration.

AZURE_OPENAI_API_KEY

Google Vertex AI

Google Cloud route via an OpenAI-compatible endpoint with access-token or service-account OAuth.

Google Cloud auth

Open-weight hosts

Groq

Open-weight models at high tokens-per-second.

GROQ_API_KEY

Cerebras

Open-weight models at high tokens-per-second.

CEREBRAS_API_KEY

DeepSeek

DeepSeek's own DeepSeek-V3 and R1 models.

DEEPSEEK_API_KEY

Together

Hosted open-weight inference (Llama, Qwen, Mixtral).

TOGETHER_API_KEY

Fireworks

Hosted open-weight inference.

FIREWORKS_API_KEY

DeepInfra

Hosted open-weight inference.

DEEPINFRA_API_KEY

Baseten

Hosted open-weight inference.

BASETEN_API_KEY

Mistral

Mistral La Plateforme hosted models.

MISTRAL_API_KEY

xAI

xAI Grok models over the OpenAI-compatible API.

XAI_API_KEY

Cloudflare Workers AI

Open-weight models served on Cloudflare's edge.

Cloudflare API token

Local & other compatible endpoints

Ollama

Local models served by Ollama. Context and availability are runtime-defined.

local runtime

LM Studio

Local models served by LM Studio's OpenAI-compatible server.

local runtime

vLLM

Self-hosted vLLM OpenAI-compatible server.

local runtime

llama.cpp

Self-hosted llama.cpp-style OpenAI-compatible server.

local runtime

Any OpenAI-compatible endpoint

Point the openai_compatible preset at any other endpoint that speaks the OpenAI Chat Completions API, including a self-hosted LiteLLM proxy.

custom base URL

Routing and configuration

Squeezy can send simple, mechanical turns to a provider's small, fast model and escalate to your main model only when a task turns out to be hard, routing never crosses providers. See how it works for the full picture.

Provider settings resolve from CLI flags, environment variables, project config, user settings, and built-in defaults. Set the provider and let the credential come from its environment variable:

[model]
provider = "openrouter"
model = ""          # set your model id, or leave blank for the default

[providers.openrouter]
api_key_env = "OPENROUTER_API_KEY"