AI Providers

Astonish supports 15+ AI providers. The easiest way to configure a provider is the interactive setup wizard:

astonish setup

You can also add and manage providers through Studio Settings > Providers.

Supported Providers

Provider	Type Key	Description
OpenAI	openai	GPT-4o, GPT-4, GPT-3.5, o1, o3
Anthropic	anthropic	Claude 4, Claude 3.5 Sonnet
Google Gemini	gemini	Gemini 2.5 Pro, Flash
AWS Bedrock	bedrock	Access Anthropic, Meta, etc. via AWS
Azure OpenAI	azure	OpenAI models via Azure
Ollama	ollama	Local models (Llama, Mistral, etc.)
OpenRouter	openrouter	Multi-provider routing
Groq	groq	Ultra-fast inference
DeepSeek	deepseek	DeepSeek models
Fireworks	fireworks	Fast inference platform
Cerebras	cerebras	Fast inference
Together	together	Open-source model hosting
Mistral	mistral	Mistral models
xAI	xai	Grok models
LM Studio	lm_studio	Local model server
LiteLLM	litellm	Universal LLM proxy
SAP AI Core	sap_ai_core	SAP enterprise AI
Poe	poe	Poe.com models
OpenAI Compatible	openai_compat	Any OpenAI-compatible API

Manual Configuration

If you prefer to edit the config file directly, providers are defined in config.yaml:

general:
  default_provider: my-openai
  default_model: gpt-4o

providers:
  my-openai:
    type: openai
    api_key: "sk-..."
    model: gpt-4o

  local-ollama:
    type: ollama
    base_url: "http://localhost:11434"
    model: llama3.1

  my-anthropic:
    type: anthropic
    api_key: "sk-ant-..."

Quick Setup

Run astonish setup for an interactive walkthrough that configures your provider and stores credentials securely. You can re-run it at any time to add more providers or change settings.

Multiple Providers

You can configure several providers and switch between them:

• Use the --provider flag on the CLI
• Change the active provider in Studio settings

API Key Security

API keys are automatically scrubbed from the config file into the encrypted credential store after initial setup. You do not need to keep plaintext keys in the config file.

OpenAI-Compatible APIs

The openai_compat type works with any OpenAI-compatible API endpoint, including vLLM, text-generation-inference, and other local or remote servers.