mnemostack 0.1.0a14

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 LLM inference layer.

Status: 🚧 alpha — API may change between 0.1.x releases.

Who is this for?

Build it in if you need:

• Long-lived agent memory that survives session restarts and doesn't drift into irrelevance as the corpus grows.
• Recall quality on mixed workloads — exact-token lookups (IDs, tickers, error strings), semantic queries, temporal questions, multi-hop reasoning — not just one of them.
• A stack you can plug into your own infrastructure: bring your own embedding model, LLM, vector store, or graph DB.

Not the best fit if you only need a single call to text-embedding-3-small + cosine similarity — something simpler will do. mnemostack earns its complexity on mixed, long-horizon workloads.

How it works, in one paragraph

On each recall(query): the four retrievers (Vector, BM25, Memgraph, Temporal) run in parallel and return ranked lists. Reciprocal Rank Fusion merges them. The 8-stage pipeline reweights results using query classification, exact-token rescue, gravity/hub dampening (to avoid always-winning popular chunks), freshness, inhibition-of-return (to not return the exact same thing twice in a row), curiosity boosts, a Q-learning reranker learned from usage, and graph resurrection (pull in related facts that weren't in top-K). An optional LLM reranker does a final ordering pass. You get a list of RecallResult with source, score, and provenance — ready to hand to a model.

Benchmarks

Full LoCoMo run (official SNAP-Research dataset, 10 samples / 1986 QA, clean state, judged by Gemini Flash):

Metric	mnemostack
Correct (strict)	66.4% (1319 / 1986)
Partial	12.8% (254)
Wrong	20.8% (413)
Combined (correct + partial)	79.2%

By question category:

Category	Correct
cat_5 adversarial open-domain	90.1%
cat_4 multi-hop reasoning	69.2%
cat_2 temporal	64.5%
cat_1 single-hop lists	34.8%
cat_3 open-domain reasoning	31.2%

Honest numbers disclaimer. The table above is our full-benchmark number across all 1986 questions and all 5 categories. Some vendors report their strongest sub-category only; if we did the same we could honestly claim 90.1% on adversarial open-domain or 69.2% on multi-hop reasoning. We publish the full aggregate because that's what actually predicts how the system behaves on mixed workloads.

How that compares with reported numbers from other systems on the same benchmark (caveat: different judges, evaluation protocols, and in some cases category cherry-picking):

System	LoCoMo correct
Hindsight (leader)	78–85%
Memobase (temporal subset)	85%
Letta filesystem agent	74%
Mem0 graph variant	~68.5%
mnemostack	66.4%
Zep (independently replicated)	58.4%

Real-corpus needle benchmark

LoCoMo measures generic long-term dialogue recall. We also run a private needle-in-haystack benchmark on the production workload that drove the original design — a ~17k-point memory stack indexed from a long-running assistant. Queries mix exact tokens (IP addresses, tickers), telegram IDs, paraphrased facts, and temporal probes.

Metric	Value
recall@1	90% (9/10)
recall@5	100% (10/10)
recall@10	100% (10/10)
Query latency p50	1.26 s
Query latency max	1.70 s

Useful because LoCoMo's failure modes (list exhaustion, open-domain reasoning) are orthogonal to what production memory stacks actually spend time on (find the specific fact the user mentioned weeks ago). This benchmark is not in the public repo; its methodology is in benchmarks/synthetic_longhorizon.py, which is the closest reproducible approximation.

Reproduce LoCoMo from a fresh clone

pip install -e '.[dev]'
bash benchmarks/download_locomo.sh   # fetches SNAP Research's public dataset
export GEMINI_API_KEY=...
bash benchmarks/run_locomo.sh        # full 10-sample run, writes results/ts.{json,log}

Details, category definitions, and notes on the judge protocol: benchmarks/README.md.

Features

• 🧠 4-source hybrid retrieval — Vector (Qdrant) + BM25 (exact tokens) + Memgraph (knowledge graph) + Temporal (time-aware vector), all fused via Reciprocal Rank Fusion. Pluggable Retriever abstraction — add your own sources.
• ⚡ 8-stage recall pipeline — ClassifyQuery → ExactTokenRescue → GravityDampen → HubDampen → FreshnessBlend → InhibitionOfReturn → CuriosityBoost → QLearningReranker. Opt-in, with persistent state store.
• 🔁 LLM reranker — Gemini Flash (or any LLM) reorders top-K by relevance; catches cases where embedding similarity alone is too broad.
• ⚡ Async-friendly — Recaller.recall_async dispatches retrievers in parallel; five concurrent HTTP recalls finish in roughly one single-recall wall-clock.
• 🌍 Unicode-aware entity resolution — Memgraph retriever probes by telegram_id, handle, and precomputed name_lower so non-ASCII names match correctly (Memgraph's toLower() lower-cases ASCII only).
• 📥 Streaming Ingestor API — batched, idempotent, LRU-cached ingest from any Python code. Same (source, offset, text) → same deterministic UUID-5 id, so re-runs are no-ops.
• 🌐 HTTP API (optional) — pip install 'mnemostack[server]' gives you /recall, /answer, /health, /docs, plus /metrics in Prometheus text format. See the HTTP server section below.
• 🔌 Pluggable embeddings — Gemini, Ollama, or HuggingFace (local GPU), via provider registry
• 🤖 Pluggable LLM — Gemini Flash / Ollama for answer generation and reranking
• 📚 Temporal knowledge graph — facts have valid_from/valid_until, query point-in-time state; graph resurrection stage recovers evicted-but-relevant memories.
• 💬 Answer mode — inference layer synthesizes concise factual answers with source citations and confidence
• ✂️ Chunkers — plain, fixed-size, and MessagePairChunker for chat transcripts (keeps user↔assistant pairs together).
• 🔎 Query expansion — optional QueryExpander rewrites short queries for better recall before fusion.
• ⚙ Consolidation runtime — phase orchestrator for nightly memory lifecycle
• 🔌 MCP server — expose memory tools to Claude Desktop, ChatGPT, Cursor, etc.
• 🛡 Graceful degradation — retrieval keeps working if graph or any retriever is down

Environment

Variable	Purpose	Required for
GEMINI_API_KEY	Google Generative AI key	Gemini embedding + Gemini Flash LLM
OLLAMA_HOST	Ollama server URL (default http://localhost:11434)	Ollama embeddings / LLM
MNEMOSTACK_COLLECTION	Qdrant collection name (default mnemostack)	CLI convenience
MNEMOSTACK_QDRANT_URL	Qdrant URL (default http://localhost:6333)	Remote Qdrant
MNEMOSTACK_GRAPH_URI	Memgraph bolt URI (default bolt://localhost:7687)	Graph retriever / GraphStore

Only the providers you actually use need their keys. HuggingFace local-GPU embeddings need no keys at all.

Try it in 30 seconds (Docker)

Fastest way to kick the tyres. No Python install, no manual Qdrant / Memgraph setup.

git clone https://github.com/udjin-labs/mnemostack && cd mnemostack
cp README.md examples/notes/              # any markdown will do
GEMINI_API_KEY=your-key docker compose -f examples/docker-compose.yml up -d --build

# Index the notes volume and ask a question over HTTP
docker compose -f examples/docker-compose.yml exec mnemostack \
    mnemostack index /data --provider gemini --collection demo

curl -s http://localhost:8000/recall \
    -H 'content-type: application/json' \
    -d '{"query":"what is this about","limit":5}' | jq

The mnemostack container runs the HTTP API on port 8000 by default. Interactive docs are at http://localhost:8000/docs. Use docker compose exec mnemostack mnemostack <cmd> for CLI-style operations (index, search, health) against the same stack.

Tear down with docker compose -f examples/docker-compose.yml down -v (the -v wipes Qdrant + Memgraph state).

Prefer Ollama (no cloud key needed)? Run Ollama on the host, set OLLAMA_HOST=http://host.docker.internal:11434, and pass --provider ollama everywhere instead of gemini.

Installation

# From PyPI
pip install mnemostack

# Optional extras
pip install 'mnemostack[huggingface]'  # local GPU embeddings
pip install 'mnemostack[mcp]'          # MCP server
pip install 'mnemostack[dev]'          # tests + linters

Run a local Qdrant for the vector store:

docker run -p 6333:6333 qdrant/qdrant:latest

Optionally a Memgraph for the knowledge graph:

docker run -p 7687:7687 memgraph/memgraph:latest

Quick start

CLI

# Health check
mnemostack health --provider ollama

# Index a directory of notes
mnemostack index ./my-notes/ --provider gemini --collection my-memory --recreate

# Hybrid recall
mnemostack search "what did we decide about auth" --provider gemini --collection my-memory

# Synthesize answer
mnemostack answer "what is the capital of France" --provider gemini --collection my-memory

# MCP server (for Claude Desktop, Cursor, etc.)
mnemostack mcp-serve --provider gemini --collection my-memory

Streaming ingest API

When you want to feed items into mnemostack from code — a chatbot that logs every message, a scraper, a daemon tailing a log — use the Ingestor. It handles batching, deduplication, and idempotency for you.

from mnemostack.embeddings import get_provider
from mnemostack.vector import VectorStore
from mnemostack import Ingestor, IngestItem

emb = get_provider("gemini")
store = VectorStore(collection="my-memory", dimension=emb.dimension)
store.ensure_collection()

ing = Ingestor(embedding=emb, vector_store=store, batch_size=64)

stats = ing.ingest([
    IngestItem(text="alice joined acme on 2024-03-01", source="notes/alice.md"),
    IngestItem(text="alice left acme on 2025-06-15", source="notes/alice.md", offset=100),
])
print(stats)  # IngestStats(seen=2, embedded=2, upserted=2, skipped=0, failed=0)

Guarantees:

• Idempotent. Each item gets a deterministic UUID5 id computed from (source, offset, text). Re-running with the same input is a no-op: Qdrant upsert replaces the point onto itself, and an in-process LRU cache skips even the embedding call for items already seen in this session.
• Batched. Items are embedded in batches of batch_size, so provider HTTP overhead amortises across many items.
• Streaming-friendly. ing.stream(item_iter) yields per-batch stats so long feeds can be monitored without waiting for the whole stream to drain.
• Graceful. If a single item fails to embed, it is counted as failed but the rest of the batch still lands.

# Long-lived feed (e.g. inside a FastAPI or Celery worker)
for item in your_firehose():
    ing.ingest_one(IngestItem(text=item.body, source=item.channel, metadata={
        "user_id": item.user_id,
        "ts": item.ts.isoformat(),
    }))

Python API

from mnemostack.embeddings import get_provider
from mnemostack.vector import VectorStore
from mnemostack.recall import Recaller, AnswerGenerator
from mnemostack.llm import get_llm

emb = get_provider("gemini")
store = VectorStore(collection="my-memory", dimension=emb.dimension)
store.ensure_collection()

# ... index data here ...

recaller = Recaller(embedding_provider=emb, vector_store=store)
results = recaller.recall("what did we decide", limit=10)

# Each result: .id .text .score .source ("vector" | "bm25" | "memgraph" | "temporal") .metadata

# Optional: synthesize a concise answer
gen = AnswerGenerator(llm=get_llm("gemini"))
answer = gen.generate("what did we decide", results)
print(answer.text, answer.confidence, answer.sources)

Full stack: 4-source retrieval + 8-stage pipeline + reranker

This is the configuration that produced the 66.4% / 79.2% LoCoMo numbers above.

from mnemostack.embeddings import get_provider
from mnemostack.llm import get_llm
from mnemostack.vector import VectorStore
from mnemostack.recall import (
    Recaller, Reranker,
    VectorRetriever, BM25Retriever,
    MemgraphRetriever, TemporalRetriever,
    build_full_pipeline,
)
from mnemostack.recall.pipeline import FileStateStore

emb = get_provider("gemini")
store = VectorStore(collection="my-memory", dimension=emb.dimension)

retrievers = [
    VectorRetriever(embedding=emb, vector_store=store),
    BM25Retriever(docs=bm25_docs),                       # see "Building a BM25 corpus" below
    MemgraphRetriever(uri="bolt://localhost:7687"),      # optional
    TemporalRetriever(embedding=emb, vector_store=store),
]
recaller = Recaller(retrievers=retrievers)
raw = recaller.recall("what did we decide", limit=30)

pipeline = build_full_pipeline(state_store=FileStateStore("/tmp/mnemo-state.json"))
reranked = pipeline.apply("what did we decide", raw)
reranker = Reranker(llm=get_llm("gemini"), max_items=20)
final = reranker.rerank("what did we decide", reranked)[:10]

Building a BM25 corpus

BM25Retriever needs a list of BM25Doc. Each doc is the atomic unit BM25 will rank — typically a paragraph or chunk of one of your source files:

from mnemostack.recall import BM25Doc
from pathlib import Path

docs = []
for i, path in enumerate(Path("my-notes/").rglob("*.md")):
    text = path.read_text()
    # chunk however you like — here: 800-char windows
    for j in range(0, len(text), 800):
        chunk = text[j : j + 800]
        if chunk.strip():
            docs.append(BM25Doc(
                id=f"{path.name}:{j}",
                text=chunk,
                payload={"source": str(path), "offset": j},
            ))

For transcript-like inputs (user↔assistant messages), prefer MessagePairChunker so a question and its answer stay in the same chunk. See mnemostack.chunking.

HTTP server (optional)

If you want mnemostack available to callers that aren't Python — any service written in Node, Go, Rust, or a plain curl from a shell script — install the server extra and expose it over HTTP:

pip install 'mnemostack[server]'
export GEMINI_API_KEY=...
mnemostack serve --provider gemini --collection memory --port 8000

Endpoints:

Method	Path	Purpose
GET	/health	Qdrant + Memgraph reachability + config summary
POST	/recall	Hybrid recall with optional 8-stage pipeline
POST	/answer	Recall + LLM answer synthesis with citations
GET	/metrics	Prometheus scrape endpoint (counters + summary histograms)
GET	/docs	Interactive OpenAPI UI

curl -s http://localhost:8000/recall \
    -H 'content-type: application/json' \
    -d '{"query": "what did we decide about auth", "limit": 10}' | jq

Response shape (abridged):

{
  "query": "what did we decide about auth",
  "results": [
    { "id": "...", "text": "...", "score": 0.72, "source": "notes/...md", "metadata": {} }
  ]
}

The /answer endpoint adds { answer, confidence, sources } alongside the memories. If the LLM isn't configured, /answer returns 503 and /recall still works — graceful degradation applies at the HTTP layer too.

For production, front this with whichever reverse proxy you already use (nginx, Caddy, Traefik) and set an auth layer — mnemostack's server does not do auth itself on purpose; the goal is to plug into whatever you already have.

Knowledge graph (optional)

from mnemostack.graph import GraphStore

graph = GraphStore(uri="bolt://localhost:7687")
graph.add_triple("alice", "works_on", "project-x", valid_from="2024-01-01")
graph.add_triple("alice", "works_on", "project-y", valid_from="2024-07-01")

# Who was alice working on in March?
march_facts = graph.query_triples(subject="alice", as_of="2024-03-15")

MCP server for Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "mnemostack": {
      "command": "mnemostack",
      "args": ["mcp-serve", "--provider", "gemini", "--collection", "my-memory"],
      "env": {
        "GEMINI_API_KEY": "your-key-here"
      }
    }
  }
}

Claude will then be able to call mnemostack_search, mnemostack_answer, and graph tools.

Custom embedding provider

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)

Install for an OpenClaw-style agent

If you run an OpenClaw-style assistant with its own scripts/recall.sh or similar recall entry point, you don't have to wire mnemostack in by hand. Ask the assistant to do it and it can follow the steps below verbatim.

(These steps are written against the markdown-backed memory layout popularised by OpenClaw / OpenClaw — the reasoning applies to any assistant whose memory lives in MEMORY.md + daily notes.)

Assistant, read this section and follow the steps if your user asks you to install mnemostack.

1. Install the package into whichever Python env your recall scripts use.

   pip install mnemostack
   

   Confirm with python3 -c "import mnemostack; print(mnemostack.__version__)".

2. Make sure Qdrant and (optionally) Memgraph are reachable. If either is missing, start them:

   docker run -d -p 6333:6333 --name qdrant qdrant/qdrant:latest
   docker run -d -p 7687:7687 --name memgraph memgraph/memgraph:latest   # optional
   

   Check with curl -sf http://localhost:6333/collections and echo 'RETURN 1;' | cypher-shell -a bolt://localhost:7687.

3. Populate the vector store with the assistant's existing memory. Almost every OpenClaw-style assistant stores memory as markdown files — MEMORY.md, memory/*.md daily notes, maybe USER.md / TOOLS.md. Those files are the ground truth; Qdrant is just a searchable index on top of them. So "migration" here really means a first-time ingest.

   Two cases:

   Case A — user has no Qdrant yet (default OpenClaw setup, markdown-only memory).

   Stand up Qdrant (see step 2), then index the workspace in one command:

   mnemostack index ~/.openclaw/workspace/ \
       --provider gemini \
       --collection memory \
       --chunk-size 800
   

   This walks every *.md / *.txt under the workspace, chunks them, embeds each chunk with the chosen provider, and upserts into Qdrant. For a typical workspace (a few megabytes of notes) this finishes in a minute or two and stays well inside free-tier embedding quotas.

   Re-running is safe. mnemostack index uses a deterministic id per (source, offset, content) chunk, so a second run skips unchanged chunks (no duplicates, no wasted embedding calls) and only re-embeds chunks whose content changed or moved. Use --recreate if you deliberately want to wipe and rebuild from scratch.

   Memgraph is optional — skip it for a pure markdown-backed memory and MemgraphRetriever just won't be part of the retriever list.

   Case B — user already has Qdrant / Memgraph populated by another plugin or a prior assistant run. Don't re-index. mnemostack reads existing collections and graphs as-is; payload like chat_id, memory_class, valid_from is preserved. Just point VectorStore(collection=...) at the existing collection name and MemgraphRetriever(uri=...) at the existing graph.

   Critical for both cases: use the same embedding model for the query that was used for the stored points. Mixing providers/models across ingest and query silently returns garbage — no error, just bad results.

4. Point your recall entry at mnemostack. Swap your search helper to use the 4-retriever Recaller from the "Full stack" example above. If your recall script has a --unified flag, route to mnemostack first and fall back to the legacy code path on any exception.

5. Keep the fallback wired. Don't delete the existing recall implementation. Route mnemostack as the primary, fall back to the old recall on any exception. A bad upgrade becomes a one-minute rollback, not an outage.

6. Smoke-test on real queries from the user's memory, not toy strings. Ask at least one exact-token query (an IP, a ticker, an error string), one semantic query, one temporal query ("what did we decide last March"), and one list query. If any regresses vs the old recall, keep both running in parallel until you understand why.

7. Write a short note in the user's daily memory describing what changed: which command was used to ingest, which embedding provider, which collection name, and where the fallback lives. Future sessions need this to avoid re-indexing with a different model and trashing recall quality.

If you're a human operator and you'd rather do this yourself, the same steps apply; treat them as a checklist.

Design

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

Pipeline state

The 8-stage pipeline needs a tiny bit of state between calls (Q-learning weights, inhibition-of-return history, per-document gravity/hub counters). FileStateStore(path) persists it to a JSON file. For multi-process servers, implement your own StateStore (two methods: load() / save(state)) backed by Redis or your database.

Graceful degradation

Any retriever can fail (Memgraph down, Qdrant unreachable, BM25 corpus empty). Recaller logs and continues with the remaining sources. The LLM reranker is wrapped in try/except by convention — if the LLM is rate-limited, the pre-rerank order is returned. This is deliberate: a memory stack that goes dark because one component hiccuped is worse than a slightly degraded one.

Roadmap

• Embedding provider registry (Gemini / Ollama / HuggingFace)
• LLM provider registry (Gemini Flash / Ollama)
• Qdrant wrapper
• BM25 + RRF recall pipeline
• Answer mode with confidence + citations
• LLM-based reranker
• Memgraph wrapper with temporal validity
• Consolidation runtime (phase orchestrator)
• CLI (mnemostack health/search/answer/index/mcp-serve)
• MCP server (Model Context Protocol)
• Text → graph triple extractor helpers (mnemostack.graph.TripleExtractor)
• Config file support YAML/JSON (mnemostack.config, mnemostack init/config CLI)
• Async variants for high-throughput servers (mnemostack.vector.AsyncQdrantStore)
• Docker compose examples (examples/docker-compose.yml)
• Reproducible LoCoMo benchmark harness in-tree (benchmarks/run_locomo.sh)
• First-class FastAPI/Starlette service wrapper (pip install 'mnemostack[server]', mnemostack serve)
• Async Recaller.recall_async and parallel retriever dispatch (proven: 5 concurrent HTTP recalls complete in ~1x single-request wall-clock)
• Benchmarks on longer-horizon synthetic corpora (benchmarks/synthetic_longhorizon.py)
• Streaming Ingestor API (mnemostack.ingest)
• Prometheus /metrics endpoint on the HTTP server
• Unicode-aware MemgraphRetriever probes (telegram_id, handle, name_lower)
• Community health: Code of Conduct, Security policy, issue/PR templates
• Per-retriever latency in /metrics (mnemostack_recall_<name>_latency_ms)

License

Apache 2.0 — see LICENSE.

Contributing

Early days. Issues and PRs welcome once API stabilizes.