Comparison

Against vault search

Literal search remains the right tool for exact strings and file discovery. Aizk is for questions that need meaning, source time, speaker perspective, shared scopes, or evidence across documents. It should complement rg and qmd rather than replace them.

Need	Vault tools	Aizk
exact term or path	fastest and simplest	unnecessary overhead
paraphrased intent	embedding search where available	dense and lexical fusion
shared project memory	no authorization model	Logto-derived scope lattice
overlap of organizations	manual duplication	native scope intersection
speaker belief or preference	prose interpretation	captured and attributed perspective
point-in-time fact replay	Git archaeology	bi-temporal range query
sourced agent context	manual note assembly	budgeted MCP context pack

Managed knowledge comparison on 2026-07-15

The comparison refreshed the full QMD index after the management notes were normalized, then asked both systems the same eight representative questions. Aizk ranked the intended current brief first for eight of eight questions. QMD without reranking did so for five of eight. This was a manual first-source check rather than an answer-generation benchmark.

Question family	Aizk first	QMD first
current open Projects	yes	yes
current Aizk state	yes	yes
current Japanese Area state	yes	no
JLPT N2 Window Weekly Plan	yes	yes
Personal Brand and Career Website	yes	yes
whether graph memory improves answers	yes	no
evaluating obsolete memory	yes	yes
next action for My Personal Computer	yes	no

QMD remained excellent when the question named one exact Project or durable note. Its failures were broad current-state questions where a related journal or Area note outranked the authority, and one conceptual paper question where a CAGRA graph-index note outranked the memory-research note. Aizk’s explicit managed-document identity, status-aware database catalogs, civil source dates, and maximal-title authority made those cases reliable without a query router.

The stable QMD path used --no-gpu --no-rerank. Its observed latency ranged from about 1.2 to 29.6 seconds and varied heavily on cold requests. CUDA initialization repeatedly failed because CMake could not resolve CUDA::cublas, while reranking sometimes stopped after announcing the rerank stage without returning results. Sequential Aizk recalls over the same representative set took about 0.7 to 2.6 seconds. The systems also return different artifacts. QMD returns matching files and snippets for an agent to interpret, while Aizk returns one ranked prompt-ready evidence string. Literal search and QMD therefore remain the better file-discovery tools, while Aizk is the better current-state memory surface in this cell.

Against published memory systems

Capability	Zep and Graphiti	Mem0	GraphRAG	Aizk
temporal facts	temporal graph	memory updates	no	valid and recorded ranges
consolidation	model-driven	add, update, delete	no	rules first, model on ambiguity
group speaker semantics	limited	user namespace	no	author snapshot and epistemic kind
authorization	application layer	application layer	no	forced PostgreSQL RLS
overlapping scopes	no	no	no	arbitrary nonempty scope sets
retrieval	graph and text	vector	community summaries	typed hybrid query and optional graph lanes
local operation	service oriented	optional	batch oriented	PostgreSQL and local model lanes

This table compares mechanisms, not scores. The honest GroupMemBench adapter now exists, but the full Aizk run has not been completed. External head-to-head claims wait for the same imported histories, answer model, judge, and hardware budget across systems.

Recent evidence also argues against assuming that more graph machinery is better. The ACL 2026 study Does Memory Need Graphs finds that raw session evidence plus independent summaries, facts, and keywords is a strong baseline. Similarity edges can add noise, and graph summaries can improve retrieval metrics while reducing answer quality if they crowd raw evidence out of the prompt. Aizk therefore keeps lane ablation and a flat baseline on the roadmap.