Graphmemo provides a sub-400ms dual-layered memory architecture (L1 Short-Term Buffer + L2 Hierarchical Graph) designed for LLM agents. It completely eliminates the latency of traditional RAG by executing graph construction asynchronously while providing instant routing.
🚀 Key Features
- < 400ms Retrieval Latency: Never wait for an LLM to read chat history. The query context is returned instantly.
- Async Graph Construction: Builds a conceptual memory graph in the background without blocking the user conversation.
- Dual Semantic Search: Prevents "Catastrophic Misrouting" by searching against both exact Topic Labels and Expanded Topic Descriptions.
- Bring Your Own Key (BYOK): Use your own free local embeddings (e.g., HuggingFace) and your own LLM provider (Groq, OpenAI, Gemini) to avoid vendor lock-in and high SaaS fees.
- 100% Local DB: Stores state cleanly in a local FAISS index + SQLite/SQLAlchemy database, meaning your data never leaves your environment.
📦 Installation
Graphmemo is lightweight and highly modular.
For a lightning-fast install (if using external API embeddings like OpenAI, Google, Anthropic):
pip install graphmemo
To include local embeddings (installs sentence-transformers and PyTorch, which may take a few minutes):
pip install "graphmemo[local]"
⚡ Quick Start
Because this library requires no paid SaaS subscription, you must inject your own LLM and Embedding functions.
import os
from pydantic import BaseModel
from typing import Optional, Any
from groq import Groq
from sentence_transformers import SentenceTransformer
from graphmemo import MemoryClient
# 1. Define your free local embedder
embedder = SentenceTransformer('all-MiniLM-L6-v2')
def embed_func(text: str) -> list[float]:
return embedder.encode(text).tolist()
# 2. Define your LLM generator (e.g., using Groq for blazing speed)
client = Groq(api_key=os.environ.get("GROQ_API_KEY"))
def llm_func(sys_prompt: str, user_prompt: str, schema: Optional[BaseModel] = None) -> Any:
# If a Pydantic schema is provided, force JSON output
# (Implementation omitted for brevity, use instructor or standard JSON mode)
pass
# 3. Initialize the Memory Client
memory = MemoryClient(
llm_generate=llm_func,
embed_text=embed_func,
use_query_expansion=True # Turn off for absolute maximum speed
)
# 4. Use it in your chatbot loop
user_id = "user_123"
# Add messages (silently batches to the Graph in the background)
memory.add_message(user_id, "user", "I bought a new Tesla Model 3 today!")
# Retrieve instant, highly accurate context for your chatbot's prompt
context = memory.retrieve_context(user_id, "What car do I drive?")
print(context['short_term_history'])
print(context['long_term_graph_context'])
print(context['global_state'])
📖 Architecture Overview
Graphmemo relies on a dual-engine system:
- The Fast Router: When a user queries your agent, Graphmemo bypasses the LLM entirely. It embeds the query and runs a highly optimized multi-dimensional FAISS search over your historical topic graph. Latency:
~300-400ms. - The Background Constructor: As the user speaks, Graphmemo silently pushes conversations into an Async Queue. An LLM agent processes batches in the background, updating topics and maintaining global state quantitative values without blocking the chat UI.
🤝 Contributing
Contributions are welcome! If you find a bug, or have a feature request, please open an issue.
📄 License
Graphmemo is MIT Licensed.
Metadata
Release files for graphmemo 0.1.2
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| graphmemo-0.1.2.tar.gz | 19.0 kB | Without provenance | Details for graphmemo-0.1.2.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| graphmemo-0.1.2-py3-none-any.whl | Python 3 | none | any | Details for graphmemo-0.1.2-py3-none-any.whl |
Total release size: 37.6 kB
Release files / graphmemo-0.1.2.tar.gz
| Download URL | graphmemo-0.1.2.tar.gz |
|---|---|
| Size | 19.0 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
2b496ab443244977bdbcc891601a0940561ba7d59073c0d2f8c6ddafd30ced82
|
|
BLAKE2b-256 checksum How to use checksums |
8279adc7fb9685aa2f95543ca8381e83afbdf803fe49275d3e6c1f7fdcd7d257
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.13.5
|
Release files / graphmemo-0.1.2-py3-none-any.whl
| Download URL | graphmemo-0.1.2-py3-none-any.whl |
|---|---|
| Size | 18.6 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
be962e675f717096af51d0f097d5bcc34fc84b85acf90bebd7822e0a3b1e89a0
|
|
BLAKE2b-256 checksum How to use checksums |
36a99f25ba0271b584af336dbc2268948e77be04b9e04456965f986e0d69b845
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.13.5
|
Security for graphmemo 0.1.2
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.