memoryweave 1.0.0

Universal long-term memory for any LLM application

MemoryWeave

Universal long-term memory for any LLM application.

---

LLMs are stateless. Every conversation starts from zero. MemoryWeave fixes that.

Plug into any LLM app with 3 lines of code. It automatically extracts entities and facts from conversations, builds a personal knowledge graph, and surfaces the most relevant context on every prompt — across sessions, users, and models.

import memoryweave

memory = memoryweave.MemoryWeave()
memory.add("My name is Ravi and I prefer Python over JavaScript.")
ctx = memory.get("What language does the user prefer?")
# inject ctx.summary into your LLM system prompt
print(ctx.summary)
# → Relevant memories:
# → - My name is Ravi and I prefer Python over JavaScript. (relevance: 0.94)

---

Features

• Model-agnostic — works with OpenAI, Anthropic, Gemini, Ollama, and any LLM
• Dual retrieval — combines semantic vector search with a structured knowledge graph
• Zero config — works out of the box with in-memory storage; swap to ChromaDB in one line
• Multi-session — isolated per-user memory with session_id
• REST API — FastAPI server so any language can use it
• TypeScript SDK — native JS/TS client for the REST API
• Fully offline — no API keys needed; runs on CPU with local models

---

Installation

pip install memoryweave

Optional extras:

pip install memoryweave[server]    # FastAPI REST server

Download the NLP model on first use:

python -m spacy download en_core_web_sm

---

Quick start

Python

from memoryweave import MemoryWeave, MemoryConfig

# in-memory store (default) — great for development
memory = MemoryWeave()

# add memories
memory.add("My name is Ravi Kashyap.")
memory.add("I work at a startup building AI tools in India.")
memory.add("I prefer Python and FastAPI for backend development.")

# retrieve relevant context
ctx = memory.get("What does this person do for work?")
print(ctx.summary)

# check stats
print(memory.stats())
# → {'session_id': 'default', 'vector_count': 3, 'node_count': 4, 'edge_count': 2}

With ChromaDB persistence

from memoryweave import MemoryWeave, MemoryConfig

memory = MemoryWeave(MemoryConfig(
    store_type="chroma",
    store_path="./my_memory_db",
    default_session_id="user-ravi",
))

memory.add("Ravi prefers dark mode and mechanical keyboards.")
ctx = memory.get("What are this user's preferences?")

With the REST API (any language)

Start the server:

uvicorn memoryweave.server:app --reload

Then from TypeScript/JavaScript:

import { MemoryWeave } from "@memoryweave/sdk";

const memory = new MemoryWeave({ sessionId: "user-123" });
await memory.add("My name is Ravi and I prefer Python.");
const ctx = await memory.get("What language does the user prefer?");
console.log(ctx.summary);

Or with plain curl:

curl -X POST http://localhost:8000/memory/add \
  -H "Content-Type: application/json" \
  -d '{"text": "Ravi prefers Python.", "session_id": "demo"}'

curl -X POST http://localhost:8000/memory/get \
  -H "Content-Type: application/json" \
  -d '{"query": "What language?", "session_id": "demo"}'

---

How it works

memory.add(text)
  │
  ├─ Extractor (spaCy)         → entities + facts
  ├─ Embedder (sentence-transformers) → 384-dim vector
  ├─ BaseStore (InMemory/Chroma)     → vector storage
  └─ KnowledgeGraph (NetworkX) → entity + fact graph

memory.get(query)
  │
  ├─ Embedder → query vector
  ├─ BaseStore.search() → top-k similar memories
  ├─ KnowledgeGraph.query() → related facts
  └─ Ranker.fuse() → weighted blend → MemoryContext

Fusion formula: score = 0.6 × vector_score + 0.4 × graph_score

---

Configuration

from memoryweave import MemoryConfig

config = MemoryConfig(
    store_type="memory",        # "memory" | "chroma" | "qdrant"
    store_path="./mw_db",       # path for chroma/qdrant
    embedding_model="all-MiniLM-L6-v2",  # any sentence-transformers model
    spacy_model="en_core_web_sm",        # any spaCy model
    top_k=5,                    # memories to retrieve per get()
    vector_weight=0.6,          # fusion weight for vector search
    graph_weight=0.4,           # fusion weight for graph search
    default_session_id="default",        # session namespace
)

---

REST API

Start the server: uvicorn memoryweave.server:app --reload

Method	Endpoint	Description
GET	/health	Health check
POST	/memory/add	Add a memory
POST	/memory/get	Retrieve context
DELETE	/memory/forget	Wipe a session
GET	/memory/stats	Session stats

Full interactive docs at http://localhost:8000/docs

---

Project status

✅ Phase 1 — Foundation
✅ Phase 2 — NLP extraction pipeline (spaCy)
✅ Phase 3 — Storage layer (vector store + knowledge graph)
✅ Phase 4 — Core memory API v0.1.0
✅ Phase 5 — TypeScript SDK
✅ Phase 6 — FastAPI REST server
⬜ Phase 7 — Documentation
⬜ Phase 8 — Launch v1.0.0 (Product Hunt + Hacker News)

Test coverage: 225+ tests · 90%+ coverage · CI green on Python 3.10/3.11/3.12

---

Contributing

See CONTRIBUTING.md for guidelines.

git clone https://github.com/ravii-k/memoryweave.git
cd memoryweave
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
python -m spacy download en_core_web_sm
pytest tests/ -v

---

License

MIT — see LICENSE for details.

---

Built by Ravi Kashyap · Started March 2026