For Developers

Three lines of code. Persistent AI memory.

Store a memory. Search for it later. That's the API. Behind those two calls, Under the hood: 3-tier cognitive memory, 5-layer hybrid search, automatic knowledge graphs, and self-improving retrieval. You focus on your agent.

$pip install hebbrix

Two ways to add memory

Hebbrix SDK

Full control over when and how memory is stored and searched.

sdk-approach.py
from hebbrix import Hebbrix

hebbrix = Hebbrix()  # Uses HEBBRIX_API_KEY env var

# Store a memory
hebbrix.memories.create(
    content="User prefers dark mode and concise responses"
)

# Search memories
results = hebbrix.search("user preferences")

# That's it. Knowledge graph, 5-layer search,
# and auto-learning happen behind the scenes.
OpenAI Drop-in

Swap your base URL. Memory search and injection is automatic.

drop-in-approach.py
import openai

# Drop-in OpenAI replacement with memory
client = openai.OpenAI(
    base_url="https://api.hebbrix.com/v1",
    api_key="your_hebbrix_key"
)

response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "What do I prefer?"}]
)
# Hebbrix auto-searches relevant memories
# and injects them before forwarding to the LLM

What you get out of the box

Every feature works automatically. No configuration, no pipelines to build.

3-tier cognitive memory

Short-term, medium-term, long-term. Memories promote based on usage and decay based on the Ebbinghaus forgetting curve.

5-layer hybrid search

Semantic vectors + BM25 + knowledge graph + importance + recency. All in under 50ms. One API call.

Automatic knowledge graph

Store natural text. Entities and relationships are extracted and mapped automatically. Zero schema definition.

Self-improving retrieval

6 automatic RL quality checks after every interaction. Good memories get reinforced. Noisy ones fade.

Collections & multi-tenancy

Isolate memories per user, team, or project. Flexible scoping that maps to your application's data model.

Natural memory decay

Old irrelevant memories don't clutter retrieval. The forgetting curve keeps your agent's context clean.

Built for developer experience

Python & TypeScript SDKs

Type-safe clients with async support

OpenAI-compatible API

Swap your base URL, keep your code

REST API

Works with any language or framework

Webhooks

Get notified on memory events

MCP integration

Works with Claude, Cline, and more

No-code connectors

Zapier and n8n integrations

Free tier

Generous limits to build and test

Full docs

Guides, reference, and examples

Start building in 5 minutes

Free tier. No credit card. Just pip install and go.