Skip to content

LLM Providers

Daystrom supports 6 provider configurations across three platforms. All use temperature: 0 and LangChain's structured output with automatic retry on parse failure.

Supported Providers

Provider SDK Auth Default Model
claude-api @langchain/anthropic API key claude-opus-4-6
claude-vertex @anthropic-ai/vertex-sdk + @langchain/anthropic GCP ADC claude-opus-4-6
claude-bedrock @langchain/aws IAM / default chain anthropic.claude-opus-4-6-v1
gemini-api @langchain/google-genai API key gemini-2.5-pro
gemini-vertex @langchain/google-vertexai GCP ADC gemini-2.5-pro
gemini-bedrock @langchain/aws IAM / default chain gemini-2.5-pro

Override the Default Model

Use --model or the LLM_MODEL env var:

# CLI flag
daystrom generate --model claude-sonnet-4-20250514

# Environment variable
export LLM_MODEL=claude-sonnet-4-20250514

Model naming differs per platform

Always use the correct format for your provider:

Platform Example
Anthropic API claude-opus-4-6
Vertex AI (Claude) claude-opus-4-6
Bedrock (Claude) anthropic.claude-opus-4-6-v1
Google AI / Vertex / Bedrock (Gemini) gemini-2.5-pro

Save Your Preference

Set your preferred provider in ~/.daystrom/config.json to avoid repeating flags:

{
  "llmProvider": "claude-api",
  "llmModel": "claude-opus-4-6"
}

Which Provider Should I Use?

Consideration SaaS APIs Vertex AI Bedrock
Setup difficulty Easiest Medium Medium
Auth mechanism API key ADC (gcloud) IAM / credential chain
Data residency No control GCP region AWS region
Billing Direct API GCP project AWS account

For detailed setup, see: