Query and index plane

Status: open/draft

Goal

One incremental vault index and one query model for the whole fleet: filters, formulas, and named views in the ergonomic shape of Obsidian Bases, served from a core index instead of per-app vault walks. Generalizes fsindex, ports vault-graph’s selector DSL, and gives agents and products the same query surface. Workload-corpus-first (per review R-8): representative real queries from vault-graph, agenda, search, polish, dashboards, and agent workflows are captured before any grammar is chosen; only index facts that corpus proves get built, and query syntax, formula semantics, the consistency model, and .base mapping are decided in a separate follow-up ADR after profiling those workloads.

Success criteria

• The query workload corpus exists first: real queries, consistency needs, update rates, and latency budgets captured from the named consumers.
• Incremental index (frontmatter, links, structural facts proven by the corpus) with file-watch invalidation; correctness under concurrent writes tested.
• Query grammar, formula semantics, and .base mapping are fixed by a follow-up ADR informed by the corpus; the API then ships in Rust, over WASM, and as CLI verbs.
• vault-graph selectors and determined-notes agenda queries run on the plane; their private walk/parse paths are deleted.
• Best-effort .base reading maps onto the native model, with documented unsupported constructs.
• At least two products (e.g. vault-search and polish) consume the shared index; their independent vault walks are gone.

Deliverables

• Index crate over the vault kernel.
• Query grammar + engine with WASM/CLI exposure.
• Migration PRs for vault-graph, determined-notes, and one more consumer.
• .base mapping notes.

Out of scope

• Obsidian Bases compatibility guarantees.
• Full-text/semantic search — vault-search’s embedding pipeline stays separate, consuming the index for its corpus walk only.

---

← Back to Milestones