Nostr RAG (Retrieval-Augmented Generation)

This module provides classes for building Retrieval-Augmented Generation (RAG) agents that can store and retrieve information from the Nostr network.

Usage

import asyncio
from agentstr import NostrRAG
# Note: To use NostrRAG, you must install the required dependencies:
# pip install "agentstr-sdk[rag]"
# You will also need an LLM API key

# Create a RAG agent, connecting to Nostr relays and using an LLM.
rag_agent = NostrRAG()

async def main():
    # Ask a question. The agent will build a knowledge base from
    # recent Nostr posts related to the query and generate an answer.
    question = "What's new with AI?"
    answer = await rag_agent.query(question, limit=5)
    print(f"Question: {question}")
    print(f"Answer: {answer}")

if __name__ == "__main__":
    asyncio.run(main())

Note

For a complete, working example, check out the Nostr RAG example.

Environment Variables

NostrRAG uses the following environment variables by default through its underlying components:

  • NOSTR_RELAYS: A comma-separated list of relay URLs to connect to. If not provided as a parameter, the agent will use this environment variable.

  • NOSTR_NSEC: The Nostr private key in ‘nsec’ format for authenticated operations. If not provided as a parameter, the agent will look for this environment variable.

  • LLM_BASE_URL: The base URL for the LLM API endpoint. If not provided as a parameter, the agent will use this environment variable.

  • LLM_API_KEY: The API key for accessing the LLM service. If not provided as a parameter, the agent will use this environment variable.

  • LLM_MODEL_NAME: The name of the LLM model to use for chat interactions. If not provided as a parameter, the agent will use this environment variable.

Note

You can override these environment variables by passing explicit parameters to the NostrRAG constructor, such as relays, private_key, llm_base_url, llm_api_key, or llm_model_name.

Reference

pydantic model agentstr.agents.nostr_rag.Author[source]

Bases: BaseModel

Show JSON schema
{
   "title": "Author",
   "type": "object",
   "properties": {
      "pubkey": {
         "title": "Pubkey",
         "type": "string"
      },
      "name": {
         "anyOf": [
            {
               "type": "string"
            },
            {
               "type": "null"
            }
         ],
         "default": null,
         "title": "Name"
      }
   },
   "required": [
      "pubkey"
   ]
}

Fields:
field pubkey: str [Required]
field name: str | None = None
class agentstr.agents.nostr_rag.NostrRAG(nostr_client: NostrClient | None = None, vector_store=None, relays: list[str] | None = None, private_key: str | None = None, nwc_str: str | None = None, embeddings=None, llm=None, llm_model_name=None, llm_base_url=None, llm_api_key=None, known_authors: list[Author] | None = None)[source]

Bases: object

Retrieval-Augmented Generation (RAG) system for Nostr events.

This class fetches Nostr events, builds a vector store knowledge base, and enables semantic search and question answering over the indexed content.

Examples

Simple question answering over recent posts:

import asyncio
from langchain_openai import ChatOpenAI
from agentstr import NostrRAG

relays = ["wss://relay.damus.io"]
rag = NostrRAG(relays=relays, llm=ChatOpenAI(model_name="gpt-3.5-turbo"))

async def main():
    answer = await rag.query(question="What's new with Bitcoin?", limit=8)
    print(answer)

asyncio.run(main())

Full runnable script: rag.py

__init__(nostr_client: NostrClient | None = None, vector_store=None, relays: list[str] | None = None, private_key: str | None = None, nwc_str: str | None = None, embeddings=None, llm=None, llm_model_name=None, llm_base_url=None, llm_api_key=None, known_authors: list[Author] | None = None)[source]

Initialize the NostrRAG system.

Parameters:

  • nostr_client – An existing NostrClient instance (optional).

  • vector_store – An existing vector store instance (optional).

  • relays – List of Nostr relay URLs (if no client provided).

  • private_key – Nostr private key in ‘nsec’ format (if no client provided).

  • nwc_str – Nostr Wallet Connect string for payments (optional).

  • embeddings – Embedding model for vectorizing documents (defaults to FakeEmbeddings with size 256).

  • llm – Language model (optional).

  • llm_model_name – Name of the language model to use (optional).

  • llm_base_url – Base URL for the language model (optional).

  • llm_api_key – API key for the language model (optional).

Raises:

ImportError – If LangChain is not installed.

async build_knowledge_base(question: str, limit: int = 10, query_type: Literal['hashtags', 'authors'] = 'hashtags') → list[dict][source]

Build a knowledge base from Nostr events relevant to the question.

Parameters:

  • question – The user’s question to guide hashtag selection

  • limit – Maximum number of posts to retrieve

Returns:

List of retrieved events

async retrieve(question: str, limit: int = 5, query_type: Literal['hashtags', 'authors'] = 'hashtags') → list[Document][source]

Retrieve relevant documents from the knowledge base.

Parameters:

  • question – The user’s question

  • limit – Maximum number of documents to retrieve

  • query_type – Type of query to use (hashtags or authors)

Returns:

List of retrieved documents

async query(question: str, limit: int = 5, query_type: Literal['hashtags', 'authors'] = 'hashtags') → str[source]

Ask a question using the knowledge base.

Parameters:

  • question – The user’s question

  • limit – Number of documents to retrieve for context

  • query_type – Type of query to use (hashtags or authors)

Returns:

The generated response

See Also