nexus-memory 1.0.1

A neuro-inspired memory architecture for AI agents — combines a Semantic Palace graph, capacity-bounded Working Memory, and asynchronous consolidation.

NEXUS Memory

A neuro-inspired long-term memory architecture for AI agents.

NEXUS combines a capacity-bounded Working Memory, a graph-based Semantic Palace, and asynchronous background consolidation to give LLM agents persistent, scalable memory — without blocking real-time interactions.

📄 Paper: NEXUS: A Scalable, Neuro-Inspired Architecture for Long-Term Event Memory in LLM Agents — Shivam Tyagi, 2025 — DOI: 10.13140/RG.2.2.25477.82407

---

Architecture

                           ┌─────────────────────────────────┐
                           │    Asynchronous Consolidation   │
                           │      (8 Background Processes)   │
                           │  • Chunking      • Cross-Ref.   │
                           │  • Conflict Res. • Skill Ext.   │
                           │  • Forgetting    • Spaced Rep.  │
                           │  • Reflection    • Defragment.  │
                           └────────────────┬────────────────┘
                                            │ background
  ┌──────────┐   ┌──────────┐   ┌───────────▼─────────┐   ┌──────────┐
  │  Input   │──▶│ Attention │──▶│   Episode Buffer    │──▶│ Semantic │
  │  Text    │   │   Gate    │   │  (append-only log)  │   │  Palace  │
  └──────────┘   │ (salience │   └─────────────────────┘   │  Graph   │
                 │  filter)  │                              │ G=(V,E)  │
                 └──────────┘                              └────┬─────┘
                                                                │
  ┌──────────┐   ┌──────────┐   ┌───────────────────┐           │
  │  Query   │──▶│ Retrieval│──▶│  Working Memory   │◀──────────┘
  │          │   │  Engine  │   │   (7 ± 2 slots)   │
  └──────────┘   │ Q(v) =   │   └───────────────────┘
                 │ β₁cos +  │
                 │ β₂decay+ │   ┌───────────────────┐
                 │ β₃freq + │──▶│    Meta-Memory    │
                 │ β₄sal    │   │ (confidence map)  │
                 └──────────┘   └───────────────────┘

Core idea: Inspired by human Dual-Process Theory (Daniel Kahneman's Thinking, Fast and Slow), NEXUS decouples memory operations into two pathways:

• System 1 (Fast & Heuristic): Real-time ingestion. Routes interactions to the short-term Episode Buffer in milliseconds without blocking the agent.
• System 2 (Slow & Analytical): Background consolidation. Uses LLM reasoning to chunk, organize, and abstract semantic knowledge asynchronously while the agent is idle.

---

Quick Start — Claude Code (MCP)

The fastest way to use NEXUS is as a persistent memory layer for Claude Code. One command, and your AI remembers you across every session.

Run the install script:

bash <(curl -s https://raw.githubusercontent.com/shivamtyagi18/nexus-memory/main/install_nexus_mcp.sh)

The script will:

• Create a dedicated venv at ~/.nexus/venv
• Install nexus-memory into it
• Prompt for your LLM choice and API key
• Register the MCP server in ~/.claude.json
• Optionally configure automatic memory hooks

Then restart Claude Code. Verify with /mcp — nexus should appear as connected.

Available tools (10):

Tool	Description
nexus_encode	Store information in long-term memory
nexus_recall	Retrieve memories by natural-language query
nexus_get_context	Inject working memory into the current prompt
nexus_how_well_do_i_know	Confidence check on a topic
nexus_knowledge_gaps	List topics NEXUS knows it doesn't know
nexus_pin	Mark a memory as permanent (never decayed)
nexus_forget	Archive a memory
nexus_consolidate	Run a consolidation cycle
nexus_stats	System-wide statistics
nexus_get_suggestions	Proactive insights from background consolidation

LLM options — set during install or via environment variables:

Model	Provider	Requires
mistral (default)	Local Ollama	ollama pull mistral
claude-*	Anthropic	NEXUS_LLM_API_KEY
gpt-*	OpenAI	NEXUS_LLM_API_KEY
gemini*	Google	NEXUS_LLM_API_KEY

---

Installation (Python Library)

pip install nexus-memory

With optional FAISS accelerated vector search:

pip install nexus-memory[faiss]

Or install from source:

git clone https://github.com/shivamtyagi18/nexus-memory.git
cd nexus-memory
pip install -e .

Prerequisites

NEXUS uses an LLM for reasoning tasks (consolidation, reflection, skill extraction). By default it connects to a local Ollama instance:

ollama pull mistral

Alternatively, you can use OpenAI, Anthropic, or Google Gemini — see Using Cloud LLM Providers below.

---

Using Cloud LLM Providers

NEXUS is provider-agnostic. Just change the llm_model and pass your API key:

from nexus import NEXUS, NexusConfig

# ── OpenAI ──────────────────────────────────────────────
config = NexusConfig(
    llm_model="gpt-4o",
    openai_api_key="sk-...",
)

# ── Anthropic ───────────────────────────────────────────
config = NexusConfig(
    llm_model="claude-3-5-sonnet-20241022",
    anthropic_api_key="sk-ant-...",
)

# ── Google Gemini ───────────────────────────────────────
config = NexusConfig(
    llm_model="gemini-1.5-flash",
    gemini_api_key="AIza...",
)

# ── Local Ollama (default) ──────────────────────────────
config = NexusConfig(
    llm_model="mistral",  # or llama3, codellama, phi3, etc.
)

memory = NEXUS(config=config)

Routing is automatic based on the model name prefix: gpt-* → OpenAI, claude* → Anthropic, gemini* → Gemini, everything else → Ollama.

---

Quick Start

from nexus import NEXUS, NexusConfig

# Initialize
config = NexusConfig(
    storage_path="./my_agent_memory",
    llm_model="mistral",
)
memory = NEXUS(config=config)

# Encode information
memory.encode("User prefers Python for backend development.")
memory.encode("User is allergic to shellfish.", context="medical")

# Recall by natural-language query
results = memory.recall("What language does the user prefer?")
for mem in results:
    print(f"  [{mem.strength:.2f}] {mem.content}")

# Check what you know (and don't know)
confidence = memory.how_well_do_i_know("programming languages")
print(f"Confidence: {confidence.overall:.0%}")

# Run background consolidation
memory.consolidate()

# Persist to disk
memory.save()

Framework Integrations

NEXUS can be used natively inside standard agent frameworks.

LangChain

Use NexusLangChainMemory to replace ConversationBufferMemory. This gives your agent the cost-savings of a capacity-bounded Working Memory while asynchronously archiving the conversation into the Semantic Palace.

from langchain.chains import ConversationChain
from nexus.integrations.langchain_memory import NexusLangChainMemory
from nexus import NEXUS

# 1. Initialize NEXUS
nexus_engine = NEXUS(storage_path="./langchain_nexus_db")

# 2. Wrap it for LangChain
nexus_memory = NexusLangChainMemory(nexus_client=nexus_engine, top_k=3)

# 3. Plug it into standard chains
conversation = ConversationChain(
    llm=my_llm,
    memory=nexus_memory,
)

conversation.predict(input="I prefer using PyTorch.")

See examples/langchain_agent.py or examples/quickstart.py for complete working code.

Claude Code (MCP Server)

See Quick Start — Claude Code (MCP) above for one-command setup.

---

Key API

Method	Description
encode(content, context, source)	Ingest new information through the Attention Gate
recall(query, top_k)	Retrieve relevant memories via graph traversal
how_well_do_i_know(topic)	Meta-memory confidence check
consolidate(depth)	Run background consolidation ("full", "light", "defer")
save()	Persist all state to disk
pin(memory_id)	Mark a memory as permanent
forget(memory_id)	Gracefully forget a memory (leaves a tombstone)
stats()	System-wide statistics

---

Configuration

All parameters are optional and have sensible defaults:

from nexus import NexusConfig

config = NexusConfig(
    # Working Memory
    working_memory_slots=7,          # Miller's Law: 7 ± 2

    # Retrieval scoring weights
    recency_weight=0.2,
    relevance_weight=0.4,
    strength_weight=0.2,
    salience_weight=0.2,

    # Forgetting
    decay_rate=0.99,                 # per-day temporal decay
    strength_hard_threshold=0.05,    # below this → forget

    # Palace graph
    room_merge_threshold=0.85,       # similarity to auto-merge rooms

    # LLM provider (pick one)
    llm_model="mistral",                     # Ollama (default)
    # llm_model="gpt-4o",                    # OpenAI
    # llm_model="claude-3-5-sonnet-20241022",# Anthropic
    # llm_model="gemini-1.5-flash",          # Google
    ollama_base_url="http://localhost:11434",

    # Storage
    storage_path="./nexus_data",
)

---

What's New in v1.0.0

• Consolidation robustness overhaul — fixed a critical bug where singleton episodes leaked in the buffer indefinitely, causing consolidation to report "no significant memories" even when important facts were present
• Smarter salience scoring — the heuristic scorer now differentiates content types (personal facts, knowledge updates, instructions) instead of scoring everything the same
• Better contradiction detection — Mistral no longer incorrectly discards memories that agree with existing ones
• Validated across 4 models — benchmarked with gpt-4o-mini, Mistral 7B, CodeLlama 7B, and Llama 3.2 3B

See CHANGELOG.md for full details.

---

Benchmarks

LoCoMo (Multi-System Comparison)

NEXUS was benchmarked against four baseline architectures on the LoCoMo long-sequence dataset (28 dialog turns, 15 evaluation questions, consolidation enabled):

System	F1 Score	Latency	Tokens/Query	Consolidation
FullContext	0.345	1147ms	550	—
MemGPT-style	0.334	1397ms	478	—
NaiveRAG	0.312	1387ms	145	—
NEXUS v2	0.279	1317ms	146	41.2s (async)
Mem0-style	0.235	1088ms	106	—

Results with GPT-4o-mini. NEXUS consolidation runs asynchronously and does not block queries.

Local Model Comparison (v1.0.0)

All runs use the fixed consolidation pipeline with heuristic scoring:

Model	F1 Score	Exact Match	Latency	Best Category
CodeLlama 7B	0.317	0.200	5634ms	Temporal (0.682)
Mistral 7B	0.284	0.067	3181ms	Knowledge Update (0.516)
gpt-4o-mini	0.262	0.000	1271ms	Single-hop (0.350)
Llama 3.2 3B	0.184	0.067	1446ms	Multi-hop (0.134)

Key finding: CodeLlama 7B outperforms all models on temporal reasoning (F1=0.682) and achieves the highest exact-match rate (20%). Mistral 7B remains the best all-rounder with strong knowledge-update handling.

LongMemEval (Long-Term Interactive Memory)

NEXUS integrates an evaluation harness for the LongMemEval benchmark to test retrieval over 50+ chat sessions:

System Configuration	Exact Match Accuracy	Average Query Latency
Baseline (Full Context)	100.0%	11.98s
NEXUS Dual-Process	80.0%	0.98s

NEXUS restricts the LLM context to the 5 most relevant memories, resulting in a >12× latency reduction compared to context-stuffing.

Vector Search Backend

NEXUS supports two vector search backends. FAISS is auto-detected when installed:

Backend	1K vectors	10K vectors	100K vectors	Memory (100K)
NumPy	22 µs	179 µs	2.75 ms	146.5 MB
FAISS	28 µs	200 µs	2.24 ms	979 B

At scale, FAISS is 1.2× faster with 150,000× less memory.

Reproducing Benchmarks

pip install -e ".[benchmarks]"

# Multi-system comparison (requires API key)
python benchmarks/run_benchmark.py --model gpt-4o-mini --systems nexus --consolidate --dataset locomo

# Local model comparison (requires Ollama)
python benchmarks/run_benchmark.py --model mistral --systems nexus --consolidate --dataset locomo
python benchmarks/run_benchmark.py --model codellama --systems nexus --consolidate --dataset locomo

# Vector backend comparison
python benchmarks/vector_benchmark.py

---

Project Structure

nexus-memory/
├── nexus/                 # Core library
│   ├── __init__.py
│   ├── core.py            # NEXUS orchestrator
│   ├── models.py          # Data models & NexusConfig
│   ├── palace.py          # Semantic Palace graph
│   ├── episode_buffer.py  # Append-only temporal log
│   ├── working_memory.py  # Capacity-bounded priority queue
│   ├── attention_gate.py  # Salience filter
│   ├── retrieval.py       # Multi-factor retrieval engine
│   ├── consolidation.py   # Async background processes
│   ├── meta_memory.py     # Confidence mapping
│   ├── vector_store.py    # Vector persistence
│   ├── llm_interface.py   # Multi-provider LLM connector (Ollama/OpenAI/Anthropic/Gemini)
│   ├── metrics.py         # Observability: counters, gauges, histograms, Prometheus export
│   └── integrations/      # Framework adapters
│       ├── langchain_memory.py  # LangChain BaseMemory component
│       └── mcp_server.py        # Claude Code MCP server (10 tools)
├── install_nexus_mcp.sh   # One-command Claude Code setup
├── tests/                 # 190 tests across 14 files
├── baselines/             # Baseline implementations for comparison
├── benchmarks/            # Benchmark harness & scripts
├── examples/              # Usage examples
├── paper/                 # IEEE research paper (LaTeX + Markdown)
│   └── figures/           # Benchmark charts and UI diagrams
├── pyproject.toml
├── CHANGELOG.md
├── LICENSE
└── README.md

---

Citation

If you use NEXUS in your research, please cite:

@article{tyagi2025nexus,
  title={NEXUS: A Scalable, Neuro-Inspired Architecture for Long-Term Event Memory in LLM Agents},
  author={Tyagi, Shivam},
  year={2025},
  doi={10.13140/RG.2.2.25477.82407}
}

---

License

MIT — see LICENSE for details.