Prompt hubs are external platforms for managing, versioning, and sharing prompts. AgentPool integrates with popular prompt management platforms to leverage curated prompt libraries and collaborative prompt development.

Overview¶

AgentPool supports integration with leading prompt hub platforms:

• PromptLayer: Comprehensive prompt management with versioning and analytics
• Langfuse: Open-source LLM engineering platform with prompt management
• Fabric: Community-driven prompt patterns and templates
• Braintrust: Enterprise prompt management with evaluation and testing

These integrations allow you to fetch prompts from these services by identifiers.

Configuration Reference¶

1
2

- type: promptlayer
  api_key: '!ENV ENV_VAR_NAME'  # API key for the PromptLayer API.

1
2
3
4
5
6
7
8
9

- type: langfuse
  secret_key: '!ENV ENV_VAR_NAME'  # Secret key for the Langfuse API.
  public_key: '!ENV ENV_VAR_NAME'  # Public key for the Langfuse API.
  host: !!python/object:pydantic.networks.HttpUrl  # Langfuse host address.
    _url: !!python/object/new:pydantic_core._pydantic_core.Url
    - https://cloud.langfuse.com/
  cache_ttl_seconds: 60  # Cache TTL for responses in seconds.
  max_retries: 2  # Maximum number of retries for failed requests.
  fetch_timeout_seconds: 20  # Timeout for fetching responses in seconds.

1

- type: fabric

1
2
3

- type: braintrust
  api_key: null  # API key for the Braintrust API. Defaults to BRAINTRUST_API_KEY env var
  project: null  # Braintrust Project name.

Configuration Notes¶

• API keys should be stored in environment variables
• Prompts from hubs can be referenced in agent system_prompts
• Hub integration provides automatic prompt versioning
• Some platforms offer additional features like prompt analytics and testing
• Prompts can be managed both locally and in external hubs