mnemostack 0.1.0a2

Memory stack for AI agents: BM25 + Qdrant + Memgraph + RRF retrieval, Gemini Flash inference layer, 8-stage recall pipeline.

mnemostack

Memory stack for AI agents — durable, structured, semantically searchable.

mnemostack is a hybrid memory system combining BM25, vector search (Qdrant), and knowledge graph (Memgraph) with a unified recall pipeline, reranker, and optional inference layer.

Status: 🚧 alpha — under active development.

Features

• 🧠 Hybrid retrieval — BM25 (exact tokens) + vector (semantic) + graph (relationships), fused via Reciprocal Rank Fusion
• 🔌 Pluggable embeddings — Gemini, Ollama, or HuggingFace (local GPU), via provider registry
• 📚 Temporal knowledge graph — facts have valid_from/valid_until so you can query point-in-time state
• 💬 Answer mode — inference layer synthesizes concise factual answers with source citations and confidence scores
• 🔄 Consolidation lifecycle — nightly decay, extraction, and promotion of memories
• 🛡 Graceful degradation — retrieval keeps working if graph is down

Installation

# Clone and install in dev mode
git clone https://github.com/YOUR_ORG/mnemostack.git
cd mnemostack
python3 -m venv .venv && source .venv/bin/activate
pip install -e .

# Optional extras
pip install -e ".[huggingface]"  # local GPU embeddings
pip install -e ".[mcp]"          # MCP server
pip install -e ".[dev]"          # tests + linters

Quick start

Pick an embedding provider

from mnemostack.embeddings import get_provider

# Option A: Gemini (cloud, best quality)
provider = get_provider("gemini")  # reads GEMINI_API_KEY from env

# Option B: Ollama (local, no API key)
provider = get_provider("ollama", model="nomic-embed-text")

# Option C: HuggingFace (local, GPU)
provider = get_provider("huggingface", model="BAAI/bge-large-en-v1.5")

vec = provider.embed("What did we decide about authentication?")
print(f"{provider.name} — dim {provider.dimension}")

Health check

ok, msg = provider.health_check()
print("provider ok" if ok else f"down: {msg}")

Custom provider

Any class that inherits EmbeddingProvider can be registered:

from mnemostack.embeddings import EmbeddingProvider, register_provider

class MyProvider(EmbeddingProvider):
    @property
    def name(self): return "my-provider"
    @property
    def dimension(self): return 512
    def embed(self, text): ...
    def embed_batch(self, texts): ...

register_provider("my-provider", MyProvider)

Design

See ARCHITECTURE.md for detailed design: pipeline stages, Qdrant schema, Memgraph temporal model, consolidation runtime, MCP tools.

Roadmap

• Embedding provider registry (Gemini / Ollama / HuggingFace)
• Qdrant wrapper + chunking
• BM25 + RRF + reranker pipeline
• Answer mode (Gemini Flash inference)
• Memgraph wrapper + temporal queries
• Consolidation runtime (decay, promote, summarize)
• CLI (mnemostack search, mnemostack index, mnemostack runtime)
• MCP server (Model Context Protocol for Claude/GPT clients)

License

Apache 2.0 — see LICENSE.

Contributing

Early days. Issues and PRs welcome once API stabilizes.