recollect-ai 0.2.0

Long-term memory layer for AI agents and assistants

Recollect

Home · Interactive Showcase · LangChain · LlamaIndex

Long-term memory layer for AI agents and assistants—self-hosted, library-first, and integration-ready for LangChain and LlamaIndex OSS workflows.

Why Recollect

• ADD-only facts from conversations (optional LLM extraction)
• Scoped memory — user_id, agent_id, run_id
• Hybrid retrieval — semantic + keyword + entity boost
• Pluggable stores — SQLite (default), Qdrant, Chroma, pgvector
• No platform required — embed in your app; open Apache 2.0

Install

git clone https://github.com/cobusgreyling/recollect.git
cd recollect
pip install -e ".[dev]"
pip install -e packages/llama-index-memory-recollect   # LlamaIndex adapter

PyPI (package name recollect-ai, import recollect):

pip install recollect-ai
pip install llama-index-memory-recollect

Extras

Extra	Purpose
openai	OpenAI LLM + embeddings
langchain	Tools + Runnable helpers
llamaindex	Core types for LlamaIndex apps
qdrant / chroma / pgvector	Vector backends
huggingface	HF sentence-transformers embedder
all	Everything

5-minute local demo (no API keys)

pip install -e .
recollect demo

from recollect import Memory, RecollectConfig

memory = Memory(RecollectConfig.local_dev())
memory.add("Prefers PostgreSQL over MySQL", user_id="alice", infer=False)
print(memory.search("database", filters={"user_id": "alice"}))

LangChain

from recollect import Memory, RecollectConfig
from recollect.integrations.langchain import create_recollect_tools, memory_runnable

memory = Memory(RecollectConfig.local_dev())
tools = create_recollect_tools(memory, user_id="alice")
ctx = memory_runnable(memory, user_id="alice").invoke({"input": "database prefs"})

Examples: examples/langchain_travel_agent.py, examples/langgraph_memory_node.py

LlamaIndex

from llama_index.memory.recollect import RecollectMemory
from recollect.config import RecollectConfig

memory = RecollectMemory.from_config(
    context={"user_id": "alice"},
    config=RecollectConfig.local_dev(),
)
# SimpleChatEngine.from_defaults(llm=llm, memory=memory)

Package: packages/llama-index-memory-recollect

API

Method	Purpose
add(...)	Extract/store memories
search(query, filters={}, top_k=5)	Hybrid retrieval
get / get_all / delete	CRUD

AsyncMemory provides the same API with async/await.

Development

make install
make test
make demo

Roadmap

• LangChain tools + Runnable
• LlamaIndex RecollectMemory
• Async API, multi-backend stores
• LlamaHub listing + upstream LlamaIndex notebook PR
• LangGraph cookbook in docs site

GitHub Pages Demo

The docs/ folder contains a fully interactive showcase (no backend) that runs the same hybrid retrieval and scoping logic live in the browser:

• Add memories with any combination of user_id / agent_id / run_id
• Real-time hybrid search (semantic + keyword + entity signals, using the same local hashing embedder approach)
• Full chat simulation: retrieve context, simulated reply, write the exchange back to memory

Once you enable GitHub Pages for this repository (Settings → Pages → Source: Deploy from a branch → main / docs), the demo will be live at:

https://<yourname>.github.io/recollect/showcase.html

Acknowledgments

Patterns draw on public agent-memory research and OSS work, including additive extraction and multi-signal retrieval popularized by Mem0. Recollect is an independent codebase.

License

Apache 2.0 — see LICENSE.