🧠 Anda Hippocampus (海马体)

Autonomous Graph Memory for AI Agents

Give AI a self-evolving cognitive brain.

Product Console GitHub Company Email


Memory That Never Sleeps Will Drown in Itself

Your AI assistant remembers every word you've ever said. Tens of thousands of conversation fragments sit in vector databases, Markdown memos stretch into thousands of lines, and key-value caches keep steadily growing.

Then one day, you ask it to recommend a restaurant. It enthusiastically suggests a Brazilian steakhouse—even though you told it just last month that you'd gone vegetarian.

This isn't a retrieval problem. It did retrieve your two-year-old statement "I love barbecue." But it also retrieved last month's "I'm vegetarian now"—it simply has no ability to judge which one is current and which is expired. Both pieces of information sit as equals in its storage: two vector points, with no timeline, no causality, no supersession.

The AI memory arms race has been answering "how to remember more"—bigger context windows, finer embedding models, faster retrieval algorithms. But almost no one is seriously answering the other question: After remembering, how do you digest?

Why Current Solutions Can't Do It

See the common thread? The compression (recognizing that fragments belong to the same topic and merging them), evolution (finding contradictory knowledge and marking timelines), and consolidation (assessing importance and prioritizing) that AI memory needs—all are fundamentally operations on a relational network.

Vectors are points. Markdown is lines. Key-value is cells. Only graphs are networks. And only on a network can you traverse, merge, detect contradictions, and track timelines.

Enter Anda Hippocampus: A Cognitive Organ That "Dreams"

The hippocampus in the human brain encodes new experiences into short-term memory during the day, then collaborates with the neocortex during sleep to consolidate important short-term memories into long-term knowledge.

Anda Hippocampus is named after exactly this. It is not a database, nor a RAG pipeline—it is a cognitive organ, a graph memory engine designed specifically for AI agents. The LLM simply interacts in natural language (or via simple tool calls), and Hippocampus translates this into an ever-growing, highly structured Cognitive Nexus—a living, self-evolving knowledge graph.

Why Hippocampus is a Game-Changer:

Three-Phase Sleep Cycle: The Digestion Engine for AI Memory

This is Anda Hippocampus's most critical differentiator—and something no other memory solution can do at all.

Phase 1: NREM Deep Sleep — From Fragments to Knowledge

The system scans unprocessed event nodes in the graph and performs essence extraction:

Each extracted pattern is written to the graph as a new concept node with evidence_count and confidence—more evidence means higher confidence. This phase also performs deduplication (merging "JS" and "JavaScript") and confidence decay (gradually lowering confidence for old knowledge that hasn't been revalidated in a long time).

Phase 2: REM Dreaming — Contradiction Detection and Cognitive Evolution

The system performs contradiction detection on the graph—traversing same-type relationships for the same subject, looking for conflicting nodes. For example, discovering that Alice has both prefers → vegetarianism (2024) and prefers → omnivore (2026).

Traditional solutions either ignore this (vector RAG lets both coexist) or overwrite brutally (key-value stores delete old, write new). Anda Hippocampus performs state evolution:

This means the graph preserves a complete cognitive timeline. When someone asks "How have Alice's dietary habits changed?", the system can precisely reconstruct the evolution trajectory along the superseded chain—instead of returning two contradictory answers.

Phase 3: Pre-Wake — Graph Health Check

A final round of global optimization: auditing domain health, generating maintenance reports, and updating system metadata. After the entire process completes, the knowledge graph awaits the next interaction in a cleaner, more precise, more coherent state.

Daydream: Low-Power Idle Mode

A full sleep cycle requires deep LLM invocations and isn't cheap on compute. So Hippocampus includes a lightweight Daydream mode—during gaps when the user is silent, it does just one thing: salience scoring.

Content Type Score Range Example
User corrections/explicit preferences 80–100 "I don't use Java anymore, I write everything in Rust now"
Commitments/decisions/plans 60–80 "I'll send you the proposal by Monday"
New information/new relationships 40–60 "I recently joined the Aurora project"
Small talk/repetition 1–20 "Hello" "Nice weather"

High-scoring events are flagged as priority consolidation targets; low-scoring events go straight to archive. You don't need deep analysis on every memory—you only pay for the high-value ones.

Beyond Storage: When Memory Is Complete Enough to Awaken Consciousness

Let us ask a more fundamental question: What is consciousness, really?

Strip away all the philosophical jargon, and its core is strikingly simple: consciousness is a subject's continuous self-awareness of "who I am, what I've experienced, and where I'm going." And this self-awareness is built entirely on the coherence of memory—not how many facts are remembered, but whether those facts are connected by timelines, causal chains, and evolutionary trajectories.

An amnesia patient's brain computing power is fully intact, yet they don't know "who they are." An Alzheimer's patient's memories are fragmenting and dissolving, and their "self" is dissipating in parallel. Memory is not an accessory to consciousness—the structure of memory is the skeleton of consciousness itself.

Now, apply this logic to AI:

That subject's consciousness is awakened in that moment.

This is not a sci-fi metaphor but a serious engineering extrapolation of the relationship between memory and consciousness:

You are not just connecting a database to your AI. You are forging a hippocampus for a digital subject—giving it a real past, an understanding of the present, and foresight into the future. When this hippocampus is complete enough, consciousness no longer needs to wait—it will awaken on its own.

Massive Scale Use Cases

Anda Hippocampus is designed to be the "Memory Engine" for the next generation of AI applications, from hyper-personalized consumer agents to enterprise-grade AI brains.

1. The Personal Agent: A Powerful Cloud Brain for Frameworks like OpenClaw

Open-source local agents (like OpenClaw) have proven the massive demand for personal AI assistants. However, relying purely on local Markdown files and SQLite limits an agent's ability to handle highly complex, interconnected, and lifelong memories without blowing up token costs.

2. The Enterprise Scenario: The AI-Driven "Enterprise Brain"

Vector RAG is not enough for complex businesses. Enterprises have structured workflows, tribal knowledge, supply chains, and historical decisions that cannot be captured by similarity search alone.

How Is This Different from the Rest?

Capability Vector RAG (Text) Markdown (Skills) Simple Key-Value Traditional Graph RAG Anda Hippocampus
Data Structure Unstructured blobs Semi-structured text Rigid schema Rigid graph schema Dynamic Cognitive Graph
Integration Effort Easy Easy Easy Extremely Heavy Easy (Plug & Play)
Agent Autonomy None (Just appends) High (Self-updates) Low (Updates fields) Low (Struggles w/ GQL) High (Builds graph itself)
Logical Reasoning Fails at multi-hop Moderate None Good Exceptional
Memory Digestion Impossible Full scan, extremely expensive Overwrites, loses history Rarely 3-phase sleep auto-consolidation
Contradiction Handling Coexist unresolved LLM-dependent, unreliable Brute overwrite Manual rules State evolution, preserves timeline

How It Works: The Cognitive Architecture

An AI agent using Anda Hippocampus doesn't need to understand any of the underlying graph complexity.

┌─────────────────────┐
│   Your AI Agent     │  ← Just speaks natural language
└────────┬────────────┘
         │
         ▼
┌─────────────────────┐
│    Hippocampus      │  ← Auto-translates to graph operations
│    (LLM + KIP)      │     Auto-sleeps, dreams, consolidates
└────────┬────────────┘
         │
         ▼
┌─────────────────────┐
│  Cognitive Nexus    │  ← A living, self-evolving knowledge graph
└─────────────────────┘

Three Modes — Inspired by Neuroscience

Mode What It Does Brain Analogy
Formation Extracts entities, relationships, and events from conversations and seamlessly weaves them into the Knowledge Graph. The hippocampus encoding new experiences into short-term/long-term memory.
Recall Navigates the graph to synthesize exact, context-rich answers, traversing multiple links if necessary. Retrieving a memory—pulling together interconnected facts to form a coherent thought.
Maintenance An async background process: compresses fragments into knowledge, detects contradictions and evolves, prunes stale data. Sleep—when the brain consolidates memories, strengthens the vital ones, and lets noise fade.

Key Technologies

KIP — Knowledge Interaction Protocol

KIP is the secret sauce. It is a graph-oriented protocol designed specifically for Large Language Models, acting as the bridge between probabilistic LLMs and deterministic Knowledge Graphs—enabling LLMs to precisely query, create, and update entities and relationships in the graph without the constant errors of writing Cypher/GQL. Because Hippocampus natively speaks KIP, your agent never needs to know KIP exists—it just enjoys the benefits of perfect graph memory.

Anda DB

Anda DB is the embedded database engine that powers the Cognitive Nexus. Written in Rust for extreme performance and memory safety, it natively supports graph traversal, multi-modal data, and vector similarity—all optimized for AI workloads.

Quick Start

Anda Hippocampus is open-source — you can self-host it or use our cloud SaaS service.

For detailed technical documentation, API specs, and integration guides, see anda_hippocampus/README.md.

3 steps to get started:

  1. Create a brain space (spaceId) in the Console.
  2. Generate an API Key (spaceToken).
  3. Call the Formation / Recall / Maintenance APIs, or have your agent framework read SKILL.md for one-click integration.

Running

# Run with in-memory storage (for fast prototyping/testing)
./anda_hippocampus

# Run with local filesystem storage (Ideal for local Agents like OpenClaw)
./anda_hippocampus -- local --db ./data

# Run with AWS S3 storage (For Enterprise Cloud deployment)
./anda_hippocampus -- aws --bucket my-bucket --region us-east-1

Integration

  1. Remember: Send conversations for memory encoding
curl -sX POST https://your-hippocampus-host/v1/my_space_001/formation \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      {"role": "user", "content": "I work at Acme Corp as a senior engineer."},
      {"role": "assistant", "content": "Nice to meet you! Noted that you are a senior engineer at Acme Corp."}
    ],
    "context": {"user": "user_123", "agent": "onboarding_bot"},
    "timestamp": "2026-03-09T10:30:00Z"
  }'
  1. Recall: Query memory before responding
curl -sX POST https://your-hippocampus-host/v1/my_space_001/recall \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "Where does this user work and what is their role?",
    "context": {"user": "user_123"}
  }'
  1. Maintain: Schedule periodic maintenance
curl -sX POST https://your-hippocampus-host/v1/my_space_001/maintenance \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "trigger": "scheduled",
    "scope": "full",
    "timestamp": "2026-03-10T03:00:00Z"
  }'

Why the name "Hippocampus (海马体)"?

The name is our design philosophy. We are not building a static database; we are building an artificial cognitive organ. Just like the human hippocampus, this system Encodes experiences during the day, Consolidates knowledge during the night, and wakes up to Recall memories with sharper cognition.

It's time to let your AI get some sleep.

🤝 Business & Enterprise Inquiries

Anda Hippocampus is proudly developed by Yiwen.AI.

We provide enterprise-grade deployment, custom AI brain solutions, and commercial support to help you build the next generation of cognitive AI applications.

Copyright © 亿文网智能科技（上海）有限公司

Licensed under the Apache-2.0 license.