The six angles answer six different questions, but they all converge on the same 2026 inflection point: Bases became a core plugin in Obsidian 1.9 (May 2025) and stabilised in 1.10 (Oct 2025) with card/list/Maps views and a public layout API [1], while Dataview has gone effectively unmaintained for over a year [2]. Every recommendation in this expedition cascades from that single fact.
The schema is the load-bearing decision. Bases reads only YAML frontmatter — it silently ignores the inline key:: value Dataview idiom [3]. So the upgrade order is: design typed-Properties first (a mandatory type: discriminator à la kepano [4], flat keys since Properties has no nested objects [5], and plural reserved keys tags/aliases/cssclasses which became mandatory in 1.9 [1]) — then route every workflow into it. Capture writes into the schema (Web Clipper Interpreter does LLM template-fill of summary/tags/author [6]; AI Tagger Universe does whole-vault batch backfill [7]). Media tracking writes into it (Media DB, Book Search). Publishing reads it (publish: true is Quartz’s ExplicitPublish filter [8]). AI plugins query it (Smart Connections, Copilot, Smart Composer).
Local-first AI is the second cross-cutting thread. Ollama + nomic-embed-text + Llama 3.1 8B (or Qwen 2.5 32B for quality) is the free backbone — it runs the AI Tagger Universe retro-fill, embeds for Smart Connections discovery, and chats through Copilot’s localhost:11434 adapter using the documented OLLAMA_ORIGINS=app://obsidian.md* workaround for CORS [9]. One subtle warning recurs: don’t double-index. Running Smart Connections + Copilot Vault QA + Smart Composer RAG simultaneously means three separate embedding indexes over the same notes [10]. Pick one chat plugin; let Smart Connections own the discovery sidebar.
Three failure modes recur across angles. First, Periodic Notes is unmaintained since 2022 — migrate to Journals before automating anything calendar-shaped [11] (the KB and capture children disagree here; Journals is the correct 2026 pick). Second, the selective-publish leak: even with publish: true, Quartz emits all non-markdown attachments — “images, voice recordings, PDFs” [12]. Keep the raw vault private and mirror only flagged content into the SSG repo via GitHub Actions. Third, don’t store media catalogues in Obsidian — Hardcover (books, free GraphQL API [13]) and Simkl (unlimited free tier vs Trakt’s new 100-item watchlist+collection cap [14]) are the master; Obsidian mirrors via Media DB or Book Search.
Migration order for an existing vault. (1) Write a master-schema note declaring every Property name + type [15]; (2) replace Periodic Notes with Journals; (3) install the official Web Clipper, retire Omnivore/MarkDownload remnants [16]; (4) stand up Ollama and run AI Tagger Universe Generate tags for vault over historical notes against a hand-curated taxonomy [17]; (5) add Smart Connections (sidebar) + exactly one of Copilot/Smart Composer (chat); (6) wire Quartz with ExplicitPublish only after the schema is stable, since publish: true is the gate.
Open questions left after all six angles. Calendar → daily-note retro-fill (Google Calendar / iCal as a passive source for “shows attended last weekend”) is not covered by any 2026 plugin in the survey — the closest match is iOS Shortcuts or a custom MCP integration via MarkusPfundstein/mcp-obsidian’s patch_content tool [18]. Apple Foundation Models SDK has no dedicated Obsidian wrapper yet despite Electron exposing the native API on macOS 15.1+ [19] — an open opportunity for a 2026 plugin author.