Skip to content

🌌 HyperspaceDB v3.1.4: True Turbo 4-Bit Lloyd-Max, 1-Bit ADC Cascades, Mem0 Drop-In & Agent Trajectories

Latest
Latest

Choose a tag to compare

View all tags
@YARlabs YARlabs released this 03 Sep 08:08
v3.1.4
c687b39

We are thrilled to announce HyperspaceDB v3.1.4 — introducing cutting-edge True Turbo 4-Bit Lloyd-Max Quantization, 1-Bit Asymmetric Distance Computation (ADC) delivering a 107× speedup with 99.9% Recall@10, the official hyperspace-memory drop-in replacement for Mem0/Zep in Python and TypeScript, and built-in Multi-Step Agent Trajectory & Lyapunov Stability Tracking! 🚀

🚀 Key Highlights in v3.1.4

1. ⚡ True Turbo 4-Bit Lloyd-Max & 1-Bit ADC Quantization (107× Speedup, 99.9% Recall)

  • True Turbo Spherical Quantization (turbo): Implemented non-linear Lloyd-Max centroid codebooks ([-2.401, ..., 2.401]) combined with random orthogonal rotation and exact $L_2$ norm preservation. Delivers 7.8× RAM compression with 96.4% Recall@10 across all metric spaces.
  • 1-Bit Rotated extreme with Asymmetric Distance Computation (ADC): Enhanced 1-bit binary quantization with vector norm scaling $|V|_2$ and full-precision query projection:
    • Single-Pass: 62.8% Recall@10 at 107× raw search speedup over float32.
    • Two-Pass Cascade ($Top\text{-}100 \to \text{Rerank}$): Achieves 99.9% Recall@10 while preserving a 15–20× net throughput boost.
  • Universal Block Quantization (medium_plus): Extended 4-bit block-wise quantization ($B=16$) to non-Euclidean geometries (Poincaré, Lorentz $H^{33}$, MRL Hybrid 801D), achieving 10.6× RAM savings with 93.6% Recall@10.

2. 🧠 hyperspace-memory: Drop-In Mem0 & Zep Replacement (Python & TS/JS)

  • 100% Mem0 API Compatibility: Migrate existing AI agents by simply replacing from mem0 import Memory with from hyperspace_memory import Memory — no prompt changes or pipeline rewrites required.
  • 100× Lower Latency (< 0.5 ms): Backed by native in-RAM MRL 129D cascades and hyperbolic indexing instead of heavy relational table lookups.
  • Zero Mandatory LLM Overhead: Direct vector + graph episodic memory operations without forcing expensive LLM calls on every memory insert.
  • 98% Storage & RAM Reduction: Native integration with extreme 1-bit ADC and turbo 4-bit quantization modes.

3. 🎯 Multi-Step Agent Trajectories & Lyapunov Stability Analysis ($\lambda$)

  • Agent Run Tracking Endpoints: Added /api/admin/runs/start, /api/admin/runs/step, and /api/admin/runs/end for tracking multi-agent execution graphs, tool calls, and step-by-step reasoning vectors.
  • Lyapunov Thought Stability Exponent ($\lambda$): Automatically computes exponential divergence rates of thought trajectories on the Poincaré disk ($H^{33}$) to detect agent hallucinations, reasoning loops, and cognitive drift in real time.
  • Interactive 3D/2D Visualizer: Added interactive trajectory viewer on /trajectory in the Hyperspace Dashboard.

4. 🛠️ Zero-Code Cognitive Memory MCP Server (mcp-hyperspace-memory)

  • Dedicated Agent Memory Server: Lightweight Model Context Protocol (MCP) server exposing 8 dedicated memory tools (memory_remember, memory_recall, memory_forget, memory_update, memory_list_sessions, memory_explore_hierarchy).
  • Zero Configuration: Simply run npx -y mcp-hyperspace-memory@latest in Cursor, Claude Desktop, Windsurf, or Antigravity to grant autonomous agents permanent, structured memory.

📄 Full Changelog

Added

  • Quantization & Math Engine:
    • Implemented QuantizationMode::Turbo with 4-bit spherical Lloyd-Max non-linear centroids and orthogonal rotation.
    • Implemented Asymmetric Distance Computation (ADC) with norm scaling for QuantizationMode::Extreme (1-bit binary).
    • Added QuantizationMode::MediumPlus block quantization ($B=16$) for non-Euclidean Lorentz and Poincaré metrics.
    • Standardized turbo, medium_plus, and extreme configuration keys across all 7 SDKs, gRPC, and REST APIs.
  • Hyperspace Memory Integrations:
    • Released hyperspace-memory Python package on PyPI (integrations/hyperspace-memory-py).
    • Released hyperspace-memory TypeScript/JS package on NPM (integrations/hyperspace-memory-ts).
    • Published mcp-hyperspace-memory MCP server on NPM (integrations/mcp-hyperspace-memory).
    • Added hyperspacedb-memory cognitive skill guide in hyperspacedb-skills.
  • Agent Trajectory & Run Telemetry:
    • Added REST API endpoints for agent run lifecycle management: /api/admin/runs/start, /api/admin/runs/step, /api/admin/runs/end.
    • Added Lyapunov exponent $\lambda$ calculation and Geodesic Trust Score estimation on Poincaré manifolds.
    • Integrated interactive trajectory visualization UI into the web dashboard (/trajectory).
  • SDKs & Ecosystem:
    • Added --dry-run and dynamic per-package version resolution to release_sdks.sh to ensure zero-skip publishing.
    • Updated C++ SDK CMake build configurations (sdks/cpp) and Dart SDK (sdks/dart).

📦 Upgrading & Installation

Cargo (Rust)

[dependencies]
hyperspace-core = "3.1.4"
hyperspace-server = "3.1.4"
hyperspace-sdk = "3.1.8"

Python SDK & Memory

# Core Vector DB Driver
pip install --upgrade hyperspacedb

# Drop-in Mem0 Cognitive Memory
pip install --upgrade hyperspace-memory

TypeScript / Node.js

# Core Vector DB SDK
npm install hyperspace-sdk-ts@latest

# Drop-in Mem0 Cognitive Memory
npm install hyperspace-memory@latest

Model Context Protocol (MCP)

{
  "mcpServers": {
    "hyperspace-memory": {
      "command": "npx",
      "args": ["-y", "mcp-hyperspace-memory@latest"],
      "env": {
        "HYPERSPACE_API_KEY": "YOUR_API_KEY",
        "MEMORY_COLLECTION": "agent_memory"
      }
    }
  }
}

🛠️ SDK & Ecosystem Matrix (v3.1.4)

SDK / Integration Version Status Highlights
Rust Core v3.1.4 ✅ Stable True Turbo 4-bit Lloyd-Max & 1-bit ADC
Python SDK (hyperspacedb) v3.1.8 ✅ Stable Standardized turbo quantization mode
TypeScript SDK (hyperspace-sdk-ts) v3.1.8 ✅ Stable Full 4-bit Turbo & Extreme ADC support
Hyperspace Memory (Python) v3.1.8 ✅ Stable Drop-in Mem0 replacement (hyperspace-memory)
Hyperspace Memory (TS/JS) v3.1.8 ✅ Stable Drop-in Mem0 replacement (hyperspace-memory)
MCP Cognitive Memory v3.1.8 ✅ Stable 8 Zero-Code Memory Tools (mcp-hyperspace-memory)
MCP Full Database Server v3.1.8 ✅ Stable 30+ CRUD, Graph & Cognitive analytics tools
n8n Community Node v3.1.48 ✅ Stable No-Code Vector DB & Memory Nodes
LangChain (Python & JS) v3.1.8 ✅ Stable Zero-RAM Sidecar Payload streaming
LlamaIndex (Python & JS) v3.1.8 ✅ Stable Zero-RAM Sidecar Payload streaming
Dart / Flutter SDK v3.1.8 ✅ Stable Mobile & Desktop client with offline sync
Go SDK v3.1.0 ✅ Stable Native gRPC client
C++ / ROS2 SDK v3.1.0 ✅ Stable Embedded robotics spatial memory
AI Agent Skills v3.1.8 ✅ Stable hyperspacedb-skills for Cursor, Claude & Windsurf

Thank you to all contributors, researchers, and node operators building the universal spatial memory for autonomous AI agents! 🌌

Assets 3
Loading