spector

# Spector

> The Zero-Overhead, Agent-Ready AI Memory Engine.

Spector is an agent-ready cognitive memory engine. It gives AI agents memory that remembers, forgets, consolidates, and forms associations across four tiers (MF-001) linked by co-activation, temporal, and entity graphs, with fused semantic and hybrid retrieval. Connect agents through the built-in MCP server, call it over REST/gRPC, drive it from the Python SDK, or embed it in the JVM. Under the hood, Project Panama (off-heap, zero GC) and the Java Vector API (AVX2/AVX-512/NEON) deliver the performance.

## Key Features

- SIMD-accelerated vector search: 88µs p50, 61K QPS
- Cognitive Memory: 4-tier memory engine implementing formal MF-001 recall algebra with ultra-low latency in-process recall
- Built-in MCP Server with 16 cognitive memory tools
- 100% off-heap Panama storage — zero garbage collection pressure
- SVASQ quantization: Float32 recall at INT8 memory sizes (99.5%+ recall)
- Hybrid search: semantic (HNSW) + keyword (BM25) with RRF fusion
- Embedded or standalone deployment
- Spring AI integration
- GPU acceleration via CUDA/Panama FFM
- Distributed mode via gRPC

## Cognitive Memory Architecture

Spector Memory implements a 4-tier cognitive architecture:
- Working Memory → Episodic → Semantic → Procedural
- Scoring based on simplified ACT-R (Anderson, 1993)
- Power-law decay (Wixted, 2004) with 12-bucket lookup
- Two-Factor strengthening (Bjork & Bjork, 1992)
- Hebbian co-activation learning
- Emotional valence (amygdala-inspired)
- Sleep consolidation (hippocampus-inspired)

## Links

- Repository: https://github.com/spectrayan/spector
- Documentation: https://spectrayan.github.io/spector/
- Quick Start: https://spectrayan.github.io/spector/getting-started/quickstart/
- MCP Server Guide: https://spectrayan.github.io/spector/sdk-usage/mcp-server/
- Cognitive Memory: https://spectrayan.github.io/spector/memory/
- Architecture: https://spectrayan.github.io/spector/architecture/overview/
- Contributing: https://github.com/spectrayan/spector/blob/main/CONTRIBUTING.md

## Technical Details

- Language: Java 25
- Build: Maven 3.9+
- License: Apache 2.0
- SIMD: AVX2 / AVX-512 / NEON via jdk.incubator.vector
- Off-heap: Project Panama FFM API
- Concurrency: Virtual Threads (Project Loom)
- Modules: 25 Maven modules
- Tests: 685+ tests
- Dependencies: Zero (JDK only)