Cognee Memory System

AI knowledge engine — a memory system in 6 lines of code.

Provenance: contributed by ClawHub user @smseow001 (originally published as cognee-memory v1.0.0, MIT-0). Translated from Chinese and adapted for this repository; the verbatim original is preserved in this file's git history.

Website: https://cognee.ai GitHub: https://github.com/topoteretes/cognee Install: pip install cognee OpenClaw plugin: @cognee/cognee-openclaw

---

Core API

The four operations

Operation	Purpose	Notes
remember	Store a memory	Persisted into the knowledge graph
recall	Query memories	Automatically routes to the best search strategy
forget	Delete memories	Remove outdated or incorrect memories
improve	Optimize learning	Continuous learning improves accuracy

---

Quick start

Python API

import cognee
import asyncio


async def main():
    # Store into the knowledge graph
    await cognee.remember("Cognee turns documents into AI memory.")

    # Store into the session cache (fast)
    await cognee.remember("User prefers detailed explanations.", session_id="chat_1")

    # Query (auto-routed)
    results = await cognee.recall("What does Cognee do?")
    for result in results:
        print(result)

    # Delete
    await cognee.forget(dataset="main_dataset")


asyncio.run(main())

CLI

cognee-cli remember "Cognee turns documents into AI memory."
cognee-cli recall "What does Cognee do?"
cognee-cli forget --all
cognee-cli -ui  # open the local UI

---

Configuration

Environment variables

# LLM API key (required)
export LLM_API_KEY="your-openai-key"

# Or use another LLM provider
# See: https://docs.cognee.ai/setup-configuration/llm-providers

# Cognee Cloud (optional)
export COGNEE_SERVICE_URL="https://your-instance.cognee.ai"
export COGNEE_API_KEY="ck_..."

---

Use cases

1. Customer-support agent

User: "My invoice issue still isn't resolved"
Cognee tracks: interaction history, failed operations, resolved cases, product history
Agent reply: "Found 2 similar billing cases resolved last month; the issue was
caused by a payment-system sync delay"

2. SQL copilot (knowledge distillation)

User: "How do I compute customer retention?"
Cognee tracks: expert SQL queries, workflow patterns, schema structure, successful implementations
Agent reply: "A senior analyst solved a similar retention query — here is their approach..."

3. Cross-session memory

# Session 1
await cognee.remember("User prefers detailed explanations", session_id="user_123")

# Session 2 (query across sessions)
results = await cognee.recall("What does the user prefer?", session_id="user_123")

---

OpenClaw plugin installation

# Install through OpenClaw's plugin manager (not plain npm):
openclaw plugins install @cognee/cognee-openclaw

# Then configure Cognee as the memory provider:
openclaw cognee setup

The plugin integrates automatically via OpenClaw's hook system:

• before_prompt_build → inject relevant memories into the prompt (auto-recall)
• after_tool_call → capture tool activity as trace entries
• llm_output → capture prompt/answer pairs into the session cache
• agent_end → sync changed memory files after each run
• session_end → bridge the session cache into the permanent knowledge graph

See the plugin README for required hook permissions (allowPromptInjection, allowConversationAccess) and configuration options.

---

vs. file-based memory

Capability	File-based memory	Cognee
Storage	Files	Vector + graph dual store
Search	Keywords	Semantic + relational
Learning	None	forget + improve
Cross-agent	Unsupported	Shared knowledge graph
Visualization	None	CLI UI

---

Deployment options

Platform	Notes
Cognee Cloud	Managed service
Modal	Serverless, GPU autoscaling
Railway	Simplified PaaS
Fly.io	Edge deployment
Render	Simple PaaS

---

Example code

Full memory loop

import cognee
import asyncio


async def memory_loop():
    # 1. Learn new knowledge
    await cognee.remember("The user is learning Python programming")
    await cognee.remember("The user prefers learning by doing")

    # 2. Query related memories
    results = await cognee.recall("What are the user's learning preferences?")

    # 3. Improve based on feedback
    await cognee.improve("Correct the misunderstanding of the user's preferences")

    # 4. Forget incorrect memories
    await cognee.forget("The incorrect assumption")


asyncio.run(memory_loop())

---

Installation status

• Python package: install cognee
• OpenClaw plugin: install @cognee/cognee-openclaw separately

---

Powered by Cognee | https://cognee.ai