mnemosynth 1.0.1

Universal AI Memory Plugin — persistent, verified, hallucination-resistant memory for any LLM, agent, or coding workflow

🧠 Mnemosynth

Cognitive Memory OS for AI

Persistent, verified, hallucination-resistant memory for any LLM, agent, or AI workflow.

---

The Problem

LLMs forget everything when a session ends. RAG helps — but raw retrieval doesn't verify, decay, or reason about what it stores. You get confident hallucinations, stale facts, and no audit trail.

Mnemosynth is a cognitive memory OS, not a vector store.

---

Install

pip install mnemosynth       # Lightweight (keyword engines)
pip install "mnemosynth[ml]" # Advanced (PyTorch + Transformers)

Zero-config. No Docker. No external databases. No API keys. Everything runs locally under ~/.mnemosynth/.

---

Three-Tier Memory Model

Inspired by how the brain actually works:

Tier	Brain Region	Backend	What It Stores
Episodic 🔵	Hippocampus	LanceDB (vector)	Events, conversations, timestamped history
Semantic 🟢	Neocortex	NetworkX (graph)	Verified facts, entity relationships
Procedural 🟠	Cerebellum	JSON registry	Tools, schemas, workflows

Memories are auto-classified on write. No manual tagging required.

---

Quickstart

Python API

from mnemosynth import Mnemosynth

brain = Mnemosynth()

# Store — auto-classified into episodic/semantic/procedural
brain.remember("User prefers Python and dark mode")
brain.remember("Yesterday we finalized the auth module")
brain.remember("Deploy with: docker compose up -d")

# Retrieve with semantic search
for r in brain.recall("What languages does the user prefer?"):
    print(f"[{r.memory_type.value}] {r.content} ({r.confidence:.0%})")

# Compressed context block for LLM injection
digest = brain.digest("Starting a new project")

# Dream: cluster episodic → promote to verified semantic facts
brain.dream()

MCP Server (Claude Desktop / Cursor / Windsurf)

{
  "mcpServers": {
    "mnemosynth": {
      "command": "mnemosynth",
      "args": ["serve"]
    }
  }
}

---

Anti-Hallucination Engine

Feature	Description
🧊 Ebbinghaus Decay	Stale memories lose confidence over time (configurable half-life)
⚔️ Contradiction Detection	Flags conflicting facts via DeBERTa NLI or keyword antonyms
💜 Sentiment Scoring	DistilBERT or keyword-based emotional valence scoring
🛡️ Immune System	Blocks prompt injections, rate limits, quarantines threats
📊 Corroboration	Repeated observations boost confidence
🔗 Belief Revision	Deprecated facts link forward to their replacements

---

Agentic Framework Adapters

First-class integrations for every major AI framework:

# CrewAI
from mnemosynth.adapters.crewai import get_crewai_tools
tools = get_crewai_tools(brain)

# LangChain / LangGraph
from mnemosynth.adapters.langchain import get_mnemosynth_tools, MnemosynthMemory

# AutoGen
from mnemosynth.adapters.autogen import get_autogen_tools

# Cross-Agent Memory Bus
from mnemosynth.adapters.memory_bus import MemoryBus
bus = MemoryBus()
bus.remember("shared fact", namespace="researcher")

Install adapter deps: pip install "mnemosynth[crewai]", "mnemosynth[langchain]", etc.

---

Causal Memory Chains

Track Decision → Reason → Outcome DAGs:

from mnemosynth.engine.causal import CausalChainEngine

engine = CausalChainEngine(brain)
engine.record_chain(
    decision="Switched from PostgreSQL to SQLite",
    reasons=["Need zero-config deployment", "Single-user workload"],
    outcome="Reduced setup time from 30min to 0",
)

# Query chains by topic
chains = engine.search("database decisions")

# Get XML digest for LLM prompts
digest = engine.get_digest("why did we switch databases?")

---

Agent-to-Agent (A2A) Protocol

Cross-agent memory sharing with visibility controls:

from mnemosynth.a2a import A2AProtocol, MemoryRequest

proto = A2AProtocol(brain)
proto.register_agent("researcher", capabilities=["search", "analyze"])

# Agent requests memories
response = proto.handle_request(MemoryRequest(agent_id="coder", query="user preferences"))

# Agent shares memories
proto.share_memory("User prefers dark mode", source_agent="researcher", visibility="shared")

---

OpenTelemetry Tracing

Full observability with zero-effort instrumentation:

from mnemosynth.telemetry import instrument_brain, get_metrics

brain = Mnemosynth()
instrument_brain(brain)  # All operations now emit traces + metrics

# Or use the pre-instrumented wrapper:
from mnemosynth.telemetry import TracedMnemosynth
brain = TracedMnemosynth()

# Check metrics
print(get_metrics())  # {remember_total: 42, recall_duration_ms_avg: 3.2, ...}

---

Production Database Backends

Scale beyond the built-in stores:

# Qdrant (vector search at scale)
from mnemosynth.stores.qdrant_store import QdrantEpisodicStore
store = QdrantEpisodicStore(url="http://localhost:6333")

# FalkorDB (graph at scale)
from mnemosynth.stores.falkordb_store import FalkorDBSemanticStore
store = FalkorDBSemanticStore(host="localhost", port=6379)

# PostgreSQL (ACID persistence)
from mnemosynth.stores.postgres_store import PostgresStore
store = PostgresStore(dsn="postgresql://user:pass@localhost/mnemosynth")

Install: pip install "mnemosynth[production]"

---

MCP Tools (8 tools)

Tool	Description
add_memory	Store with auto-classification
search_memory	Semantic search across all tiers
get_digest	Compressed XML context block
get_contradictions	Surface conflicting facts
run_dream	Trigger consolidation
forget	Delete by ID
get_stats	Memory statistics
get_provenance	Full audit trail

---

CLI

mnemosynth serve              # Start MCP server
mnemosynth stats              # Memory dashboard
mnemosynth search "query"     # Semantic search
mnemosynth inspect            # Browse memory tree
mnemosynth dream              # Run consolidation
mnemosynth health             # System diagnostics
mnemosynth export -o out.json # Export to JSON
mnemosynth forget <ID>        # Delete a memory
mnemosynth reset --confirm    # Wipe everything

---

Configuration

Override defaults at ~/.mnemosynth/config.yaml:

embedding_model: all-MiniLM-L6-v2
max_episodic_memories: 10000
max_semantic_nodes: 5000

decay:
  half_life_days: 30.0
  min_confidence: 0.1

dream:
  interval_hours: 24
  min_cluster_size: 3

digest:
  max_tokens: 150
  top_k: 5

---

Architecture

Claude Desktop / Cursor / Windsurf / Python API / CrewAI / LangChain
                    |  MCP (stdio) / Python import / A2A Protocol
           Router · Digest · Decay · Dream · Causal Chains
           Sentiment · Contradiction · Immune · Telemetry
                    |
   Episodic (LanceDB/Qdrant) · Semantic (NetworkX/FalkorDB) · Procedural (JSON)
                    |
              ~/.mnemosynth/  (or PostgreSQL)

---

Optional Dependencies

Extra	What It Adds
mnemosynth[ml]	PyTorch, Transformers (DeBERTa, DistilBERT)
mnemosynth[production]	Qdrant, FalkorDB, PostgreSQL, OpenTelemetry
mnemosynth[crewai]	CrewAI adapter
mnemosynth[langchain]	LangChain adapter
mnemosynth[autogen]	AutoGen adapter
mnemosynth[adapters]	All framework adapters

---

🧠 MNEMOSYNTH — Because AI shouldn't have amnesia.
_{Built by Vasudev Jaiswal · Apache 2.0}