LlamaIndex Integration

Use NeuralSnap as a structured knowledge layer alongside LlamaIndex's RAG pipelines. Instead of flat document chunks, your index returns typed, confidence-scored Neural Snapshots with connections and context.

NeuralSnap complements your vector store — it provides structured knowledge (beliefs, decisions, frameworks) while your vector store handles unstructured document retrieval.

Installation

bash

pip install neuralsnap-llamaindex

Quick Start

python

from neuralsnap_llamaindex import NeuralSnapRetriever

# Create a retriever that searches your NeuralSnap brain
retriever = NeuralSnapRetriever(
    api_key="ns_your_api_key",
    brain_id="brain_abc123",
    top_k=5,
)

# Retrieve relevant snapshots
nodes = retriever.retrieve("how to approach complex problems")
for node in nodes:
    print(f"[{node.score:.2f}] {node.metadata['name']}")
    print(f"  Type: {node.metadata['type']}")
    print(f"  Core: {node.text}")

With a Query Engine

Combine NeuralSnap retrieval with LlamaIndex's query engine for RAG that understands structured knowledge:

python

from llama_index.core import VectorStoreIndex
from llama_index.core.query_engine import RetrieverQueryEngine
from neuralsnap_llamaindex import NeuralSnapRetriever

# NeuralSnap retriever for structured knowledge
ns_retriever = NeuralSnapRetriever(
    api_key="ns_...",
    brain_id="brain_abc123",
    top_k=5,
    include_metadata=True,  # include type, confidence, tags
)

# Build a query engine
query_engine = RetrieverQueryEngine.from_args(
    retriever=ns_retriever,
)

# Query with full context
response = query_engine.query(
    "What do we believe about scaling engineering teams?"
)
print(response)
# → Response includes snapshot type, confidence scores, and connections

Ingesting Documents

python

from neuralsnap_llamaindex import NeuralSnapIngester

# Ingest documents into NeuralSnap
ingester = NeuralSnapIngester(
    api_key="ns_...",
    brain_id="brain_abc123",
)

# From a LlamaIndex document
from llama_index.core import Document

doc = Document(
    text="In our Q4 review, we decided to focus on enterprise...",
    metadata={"source": "q4-review", "date": "2025-01-15"},
)

# Crystallize into Neural Snapshots
result = ingester.ingest(doc)
print(f"Created {len(result.snapshots)} snapshots")

Configuration

Property	Type	Description
`api_key`*	string	Your NeuralSnap API key.
`brain_id`	string	Scope to a specific brain.
`top_k`	number	Number of results to return. Default: 5.
`include_metadata`	boolean	Include snapshot type, confidence, tags in node metadata. Default: true.
`min_confidence`	number	Filter results below this confidence threshold (0–1). Default: 0.
`snapshot_types`	list[str]	Filter by snapshot type (Belief, Model, Rule, Conviction, Principle).

Early access

The LlamaIndex integration is in early access (v0.1.0). The retriever and ingester are fully typed but connect to the NeuralSnap API in v0.2.0. Install now to lock in the API surface.

CrewAI

JavaScript / Node