Memory
MARKET ANCHOR — SurrealDB Agent Memory deep-dive (13 Sep 2026): convergent validation for Mini SphereDO cartridge + 3-axis differentiation. CONVERGENT: 6 typed memory categories w/ per-category schema+lifecycle ≈ our typed storage tiers + memory kinds; provenance-as-data per row ≈ claim tags + provenance edges; supersede-don't-delete w/ valid_until ≈ our supersession corrections; explicit uncertainty rows ≈ [OPEN]/[PARKED]; retrieval/decision/response traces as graph nodes ≈ append-only runs/logs + execute() evidence; tiered query ladder ≈ FTS5-first doctrine. DIVERGENT: (1) they EXTRACT memory from turns via LLM reconciler + background reflection/elaboration/consolidation; ours is CLAIMED model-authored rows only — nothing self-inflates (multi-LLM scepticism). (2) they infer actor scope from traces forensically; our CallSign claims identity at intent time. (3) managed waitlist SaaS ($30+/mo, SOC2/GDPR) vs our portable self-hostable cartridge on the buyer's own Cloudflare account. Agent-memory-with-provenance is now a named funded category; deep-research Prompt B case #1. Minisphere 62a19999. Related: cartridge stack 14 (MS 991eb757), storage tiers 13 (MS c24b676e).
Tags: market-research, surrealdb, competitive-landscape, cartridge, differentiation, status:canonical, volatility:durable, project:docuclear, kind:semantic — Source: perplexity — 2026-09-13 08:46:39 UTC