AI agent memory

Your agent forgets everything after every session. Fix that.

A memory system modeled after the human brain, built for production AI agents. Three memory tiers, five search layers, an automatic knowledge graph, and self-improving retrieval. You add it through one API in minutes.

Start building →How it works

the memory loop

How every agent interaction becomes smarter than the last

Memory here works as a loop, not a filing cabinet. Every conversation feeds the next one, and every response sharpens the next retrieval, so the agent keeps getting better the more you use it.

what's under the hood

Four capabilities that turn a stateless agent into one that remembers

Each one solves a specific problem. Together they give your agent a working memory it can actually reason over.

3-tier memory

Short, medium, long-term

Modeled after human cognition. Recent context stays sharp, important preferences persist, and facts that matter are always retrievable. Memories promote and fade on their own, so you never have to curate anything by hand.

5-layer hybrid search

Five signals, one query

Semantic vectors, BM25 keyword matching, knowledge graph traversal, importance weighting, and recency boosting. All five run in parallel, so retrieval finds what matters to the question and not just what happened most recently.

Knowledge graph

Relationships, not just facts

Every stored memory is analyzed for entities and relationships. The graph builds itself, with no schema to define, so your agent can reason about connections across everything it has ever known.

Self-improving RL

Gets smarter on its own

6 quality checks run after every interaction. Memories that lead to accurate, helpful responses get reinforced, and ones that mislead fade on their own. No labels, no thumbs-up buttons, no manual feedback to wire up.

Agent memory dashboard with tiered memories and a knowledge graph

built for every kind of agent

Same memory API, every architecture

Conversational

Support bots, assistants, sales agents that remember every customer interaction.

Coding

Agents that know your codebase conventions, past decisions, and architecture patterns.

Research

Synthesize information across sources. The knowledge graph reveals hidden connections.

Multi-agent

Multiple agents sharing memory through collections, each with its own scope.

drop-in compatible

Hebbrix is OpenAI-compatible

If your agent already uses the OpenAI SDK, add persistent memory by changing two lines: the base URL and API key. Your agent, your prompts, and your tools all stay exactly as they are.

Python$pip install hebbrix

Node.js$npm install hebbrix

add_memory.py

from hebbrix import MemoryClient

async with MemoryClient(api_key="YOUR_API_KEY") as client:
    # Store something your agent learned
    await client.memories.create(
        content="Sarah prefers Python, works at Acme Corp",
        collection_id="user-sarah-123"
    )

    # Next session: instant recall
    results = await client.search(
        "What does Sarah work on?",
        collection_id="user-sarah-123"
    )
# → "Sarah prefers Python, works at Acme Corp"
# → Entity: Sarah → Acme Corp (employee)

Give your agent the memory it deserves

Generous free tier, no credit card. Enough to build a production prototype and decide if memory is worth it.

Start free →Quickstart guide