LLM Wiki — Synaptic Lattice

Competitive Brief — Synaptic Lattice README vs Top LLM-Wiki Competitors

Properties1
tagsquery, comparison, llm-wiki, second-brain, architecture, research, retrieval

Leitura de 4 min

Competitive Brief — Synaptic Lattice README vs Top LLM-Wiki Competitors

Question

How does this vault’s README feature-set compare to the most popular LLM Wiki implementations, where are we ahead and behind, and what should we focus on now?

Answer

The vault is ahead on discipline, provenance, and testing; behind on product surface (UI, browser capture, ingest formats, autonomous maintenance, packaging). No major competitor targets a non-technical owner with a self-hosted, governance-heavy package — that position is unclaimed.

Competitors analyzed (live GitHub READMEs, 2026-08-01):

ProjectStarsPositioning
nashsu-llm-wiki15.7kFull cross-platform desktop app (Tauri v2)
claude-obsidian10.2kLocal-first Obsidian + Claude Code, 15 skills
atomicstrata-llm-wiki-compiler1.9kCompiler/SDK + Configurable Lifecycle Profiles
lucasastorian-llmwiki1.4kWeb app + MCP + Chrome clipper, hosted/local

Where we are ahead

  1. Deterministic query + routing pipeline — route-classifier.ps1 (main/prompt wikis) plus hybrid search (BM25 + vector + graph PPR via RRF). No competitor routes queries between two governed wikis.
  2. External-link capture + freshness ledger (scripts/external_links.py) — turns URLs into immutable, content-hashed, status-tracked evidence (Guardrail #16). None of the four competitors do this.
  3. Lint-gated, review-gated ingest — wikilint.ps1 (scoped, deterministic) plus batch --claim/--decide approval before any page lands. atomicstrata is closest; nashsu/claude-obsidian do not gate writes this hard.
  4. OKF compliance + supersession/aliases/contested fields as a standard — stronger than atomicstrata’s export/import-only OKF support.
  5. Non-technical-owner focus (dashboard.ps1, plain-language docs, smoke test) — explicitly claimed; nobody else owns it.

Where we are behind

  1. No product surface — no desktop app, web UI, or interactive graph with Louvain community detection and “surprising connections” insights (nashsu owns this).
  2. No browser capture loop — nashsu/lucasastorian close capture→wiki; our external_links.py is a manual CLI.
  3. Narrow ingest formats — PDF/DOCX/PPTX/XLSX/EPUB/image ingest is table stakes for the desktop players; we are markdown/text-only.
  4. No autonomous maintenance — lucasastorian’s nightly Claude Routine and nashsu’s queue auto-watch remove the “remember to ingest” burden.
  5. No packaging/distribution — no Docker, installer, SDK, hosted mode, or binary (already noted in ecossistema-stacks-vs-nosso-vault).
  6. Lifecycle is cosmetic, not enforced — lifecycle: l0-l3 exists in frontmatter but nothing gates transitions; atomicstrata enforces them via runtime profiles.

This week: mature dashboard.ps1 into a guided no-terminal flow; wire external_links.py capture into it as “Paste URL → capture → queue”.

Next 1–3 weeks: a deterministic query gate (route-classifier + hybrid search

  • citation check as one testable artifact); formalize lifecycle transitions as wikilint gates (borrow atomicstrata’s CLP idea in OKF terms).

Strategic: packaging (Docker + installer); graph-insights layer (Louvain + isolated/bridge detection) on top of the existing graph-export.

Don’t chase: i18n, multi-format OCR, Rust backend, hosted cloud tier — they conflict with the local-trust, no-lock-in positioning the README promises.

Critique

The “Recommended focus” section of this brief was audited the same day it was filed (see audit-2026-08-01-roadmap-and-dashboard-plan). The audit’s central challenge: the recommended focus is a PM’s guess, not a validated user need. The owner-dashboard item — claimed as the NOW priority — is built on an untested assumption about who the non-technical owner is and whether the one-line CLI capture actually fails them. The audit recommends a 14-day capture-usage log before any dashboard build, and points out that a browser bookmarklet reusing external_links.py capture may serve the job with fewer steps than a 7-option menu.

This brief’s competitor facts (star counts, feature matrix, ahead/behind judgments) are not contested. Only the priority ordering is: this brief places the dashboard first; the audit places observational testing and the deterministic query gate first.

Implications

  • Treat the brief’s “Recommended focus” as a hypothesis to validate, not a committed backlog. The fact base survives; the ordering does not.
  • Any future reader deciding what to build next should weigh both positions — the brief’s dashboard-first framing and the audit’s test-first challenge — before converging on the roadmap.

Sources

^[wiki/entities/nashsu-llm-wiki.md] ^[wiki/entities/claude-obsidian.md] ^[wiki/entities/atomicstrata-llm-wiki-compiler.md] ^[wiki/entities/lucasastorian-llmwiki.md] ^[wiki/queries/ecossistema-stacks-vs-nosso-vault.md] ^[wiki/queries/quality-comparison-analysis.md]

External references (consulted 2026-08-01; references, not vault evidence per Guardrail #16): https://github.com/nashsu/llm_wiki · https://github.com/AgriciDaniel/claude-obsidian · https://github.com/atomicstrata/llm-wiki-compiler · https://github.com/lucasastorian/llmwiki

Visão de gráfico

Backlinks