🧠 OpenClaw Cortex

Hybrid semantic memory system for AI agents — vector search, smart capture, multi-factor recall.

What is it?

OpenClaw Cortex gives AI agents persistent, semantically searchable memory across sessions. It bridges file-based structured memory (markdown, git-backed) with vector-based semantic memory powered by Qdrant.

Features

Semantic Search — Vector similarity search via Qdrant gRPC
Smart Capture — Claude Haiku extracts structured memories from conversations
Multi-Factor Recall — Scores by similarity, recency, frequency, type, and scope
Token Budgeting — Fit recalled memories within configurable limits
Auto-Classification — Rules, facts, episodes, procedures, preferences
Lifecycle Management — TTL expiry, session decay, consolidation
OpenClaw Integration — Pre/post-turn hooks for seamless agent memory

Tech Stack

Go 1.25+ Qdrant Ollama Claude Haiku Cobra CLI gRPC Docker Kubernetes

Quick Start

task docker:up                # Start Qdrant
ollama pull nomic-embed-text  # Pull embedding model
task build                    # Build binary

openclaw-cortex search "how to deploy"
openclaw-cortex recall "database question" --budget 2000

View on GitHub →

Documentation

README — Full usage, config, CLI reference
Architecture — System design & data flows
Deployment — Docker, K8s, configuration
Contributing — Development setup & workflow