LLM Adapter

The LLM adapter configures which language model your agent uses. It lives at adapters/llm.yaml and is referenced by the manifest’s spec.runtime.llm.adapter field.

Configuration

# adapters/llm.yaml
provider: gemini
model: gemini-2.0-flash
auth:
  env_var: GOOGLE_API_KEY
parameters:
  temperature: 0.7
  max_tokens: 4096

Fields

Supported providers

Gemini (AI Studio)

provider: gemini
model: gemini-2.0-flash
auth:
  env_var: GOOGLE_API_KEY

Get your API key at aistudio.google.com/apikey .

Available models: gemini-2.0-flash, gemini-2.5-pro, gemini-2.5-flash

Anthropic

provider: anthropic
model: claude-sonnet-4-20250514
auth:
  env_var: ANTHROPIC_API_KEY

Available models: claude-sonnet-4-20250514, claude-haiku-4-20250414, claude-opus-4-20250514

OpenAI

provider: openai
model: gpt-4o
auth:
  env_var: OPENAI_API_KEY

Available models: gpt-4o, gpt-4o-mini, gpt-4-turbo

Vertex AI (Google Cloud)

provider: vertex-ai
model: gemini-2.0-flash
auth:
  env_var: GOOGLE_APPLICATION_CREDENTIALS
  # Required for Vertex AI
  project_env_var: GOOGLE_CLOUD_PROJECT
  location_env_var: GOOGLE_CLOUD_LOCATION

Vertex AI requires a GCP service account JSON file and project configuration. Set these in your .env:

GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json
GOOGLE_CLOUD_PROJECT=your-project-id
GOOGLE_CLOUD_LOCATION=us-central1

Available models: gemini-2.0-flash, gemini-2.5-pro

Embedding adapter

The embedding adapter follows the same pattern at adapters/embed.yaml:

# adapters/embed.yaml
provider: openai
model: text-embedding-3-small
auth:
  env_var: OPENAI_API_KEY

Vertex AI embeddings

provider: vertex-ai
model: text-embedding-005
auth:
  env_var: GOOGLE_APPLICATION_CREDENTIALS

Runtime reference

The manifest references adapters via relative paths:

# manifest.yaml (excerpt)
spec:
  runtime:
    llm:
      adapter: ./adapters/llm.yaml
    embedding:
      adapter: ./adapters/embed.yaml

zil validate checks that these files exist and are valid YAML.

How the SDK uses adapters

When you call zil.create_agent(), the SDK reads adapters/llm.yaml and maps it to an ADK-compatible model string:

Gemini and Vertex AI models are native to ADK and used directly. Anthropic and OpenAI models use the LiteLLM  prefix convention.

You can override the model at runtime:

import zil
 
# Use a different model without changing llm.yaml
agent = zil.create_agent(model="openai/gpt-4o-mini")

See SDK Reference for all options.