Skip to main content
Integrate Moss semantic search into DSPy as a retrieval module. This setup lets you use Moss in DSPy programs, including ReAct agents, with sub-10ms retrieval latency for fast, composable LLM pipelines.
Note: For a complete Jupyter notebook walkthrough, see the DSPy cookbook.

​
Why use Moss with DSPy?

DSPy’s retrieval modules connect external knowledge sources to composable LLM programs. Moss delivers sub-10ms semantic search through a standard Retrieve interface, giving your DSPy pipelines fast, accurate retrieval without managing vector database infrastructure.

​
Required tools

  • Moss account with project credentials
  • OpenAI API key (for LLM usage)
  • Python 3.10+

​
Integration guide

1

Installation

pip install inferedge-moss dspy nest-asyncio python-dotenv
2

Environment setup

Create a .env file in your project root.
.env
# Moss Credentials
MOSS_PROJECT_ID=your_project_id
MOSS_PROJECT_KEY=your_project_key
MOSS_INDEX_NAME=your_index_name

# OpenAI
OPENAI_API_KEY=sk-...
3

Use as a configured retriever

The cookbook provides a MossRM class that extends dspy.Retrieve. Register it as DSPy’s default retriever, then use dspy.Retrieve throughout your programs.
from inferedge_moss import MossClient
from moss_dspy import MossRM
import dspy

client = MossClient("your-project-id", "your-project-key")

retriever_model = MossRM(
    index_name="your-index-name",
    moss_client=client,
    k=3,
    alpha=0.5,
)

dspy.configure(rm=retriever_model)

# Now dspy.Retrieve will use Moss
retrieve = dspy.Retrieve(k=3)
results = retrieve("What is the return policy?")
print(results.passages)
4

Use as a ReAct agent tool

You can also use Moss as a tool inside a DSPy ReAct agent for multi-step reasoning with search.
from inferedge_moss import MossClient, QueryOptions
import asyncio
import nest_asyncio
import dspy

nest_asyncio.apply()

client = MossClient("your-project-id", "your-project-key")
INDEX_NAME = "your-index-name"
asyncio.get_event_loop().run_until_complete(client.load_index(INDEX_NAME))

def moss_search(query: str):
    """Searches the Moss knowledge base for relevant passages."""
    loop = asyncio.get_event_loop()
    options = QueryOptions(top_k=5, alpha=0)
    results = loop.run_until_complete(client.query(INDEX_NAME, query, options=options))
    if not results.docs:
        return "No relevant info found."
    return "\n".join([f"- {doc.text}" for doc in results.docs])

react_agent = dspy.ReAct(
    signature="question -> answer",
    tools=[moss_search],
    max_iters=5,
)

result = react_agent(question="What payment methods are accepted?")
print(result.answer)

​
Configuration

​
MossRM

ParameterTypeDefaultDescription
index_namestrRequiredThe name of the Moss index to query.
moss_clientMossClientRequiredAn initialized Moss client instance.
kint3Default number of passages to retrieve.
alphafloat0.5Hybrid search weighting. 0.0 = keyword only, 1.0 = semantic only.