How We Built an AI Agent That Never Forgets

HyperNexus implements a dual-tier memory architecture:

L1 - Session Scratchpad: Ephemeral, lightning-fast memory tied directly to the active session.

L2 - The Vault: Permanent semantic storage in SQLite with sqlite-vec for vector search. Saves exact transcripts and LLM-compressed heuristics.

Context Harvesting: Every session autonomously queries the L2 Vault to pull in relevant historical heuristics.

14,726 memories currently stored, all surviving restarts.

CREATE TABLE memories (
    id INTEGER PRIMARY KEY,
    content TEXT,
    embedding BLOB,
    tags TEXT[],
    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);

Try it: https://hypernexus.site/blog/ai-agent-memory.html

Originally published on hypernexus.site