infinity-neural — Zero-decay neural memory with instinct-level recall

Neurons never die. V3.1 adds instinct-level auto-remember and 3-layer lazy recall so agents write and retrieve memories the way humans do — automatically, without being told.

Overview

Infinity Neural V3.1 is a complete neural memory system for OpenClaw agents. V1 established zero-decay storage with 3-tier compression (HOT/WARM/COLD), ETERNAL/ANCHOR neuron types, and Vietnamese NLP. V2 added 7 engine patterns: crash sentinel, scored observations, coordination ledger, agent lock, tier decay, fact supersession, and size-based triggers. V3 installs instinct-level behavior: auto-remember 7 trigger types, 3-layer lazy recall, session-aware loading, duplicate prevention, and memory hygiene. Ships with 46 MCP tools and a MEMORY.md↔neural graph bridge.

Features

• Zero decay — no pruning, no forgetting; neurons compress but never die
• 3-tier — HOT (active) / WARM (stable) / COLD (archived)
• Auto-Remember — 7 triggers: decision, workflow, fact, insight, instruction, context, reference
• 3-Layer Lazy Recall — Tier 1 (~200 tokens) → Tier 2 (~500) → Tier 3 (~1000) on miss
• Fact supersession — correct stale facts without losing history
• 46 MCP tools — 38 core + 8 cognitive (hypothesize, predict, verify, gaps, calibrate...)
• Memory Bridge — scripts/memory_bridge.py syncs MEMORY.md ↔ neural graph
• Tested — 66 unit + 3217 engine tests PASS, schema v21

Trigger Keywords (OpenClaw)

infinity neural, neural memory, auto remember, auto recall, lazy recall, crash sentinel, fact supersession, agent coordination, bộ nhớ vĩnh cửu

Related Skills

• core-brain-mula — zero-decay neural memory setup
• memory-tiering — 3-tier memory management

---

Part of the NachaFromMars OpenClaw skill ecosystem.