Graphite Python SDK

Native Python bindings for the Graphite embedded GraphRAG engine

Graphite DB combines Knowledge Graphs, SIMD AVX2 Vector Search, BM25 Lexical Indexing, and Token-Budgeted Context Synthesis into a single-file .graphite database with zero-copy memory mapping.

Installation

pip install graphite-rag

Quickstart

import graphite_db as graphite

# 1. Open or create an embedded database
db = graphite.open("knowledge.graphite", dim=384, max_tokens=400)

# 2. Insert entities and create connections
id_auth = db.insert("AuthService", entity_type="Module", description="Validates JWT tokens")
id_db = db.insert("UsersDB", entity_type="Database", description="PostgreSQL primary cluster")
db.connect("AuthService", "UsersDB", relation="CONNECTS_TO", weight=0.95)
db.flush()

# 3. Query GraphRAG context with automatic local FastEmbed embedding
result = db.query("How does authentication connect to the database?")

print(f"Retrieved {result.token_count} tokens:")
print(result.markdown)

In-Memory Ephemeral Storage

import graphite_db as graphite

with graphite.in_memory(dim=384) as db:
    db.remember("User prefers concise Portuguese responses.", category="UserPreference")
    result = db.query("What language does the user prefer?")
    print(result.markdown)

Direct Vector Operations & Custom Embeddings

import graphite_db as graphite

# Generate 384-dimensional vector locally on CPU
vector = graphite.embed("How does authentication work?")

db = graphite.in_memory(dim=384)
result = db.retrieve_context(vector, query_text="authentication", max_tokens=300)
print(result.markdown)

License

Dual-licensed under MIT or Apache-2.0.