[Searchable]

Summary#

[Searchable] marks a String or Markdown property as searchable. It gives the field the best relevance search the app can offer: full-text always works (no setup), and semantic ranking upgrades it the moment an embedding provider is configured — with no change to your code. Two optional arguments choose where the searchable content lives (scope) and which kinds of relevance you get (mode). Requires using Memory;.

Signature#

using Osyrin.Memory;

[Searchable]                      // scope defaults by type; Full
[Searchable(Entity)]              // entity-local field search; Full
[Searchable(Memory)]             // chunked into the shared corpus; Full
[Searchable(Entity, TextOnly)]   // entity-local; lexical only (no embedder needed)

Valid only on a String or Markdown property. Enums: SearchScope { Entity, Memory }, SearchMode { Full, TextOnly }. Both arguments are optional and order-independent.

Description#

A [Searchable] field participates in relevance search. What you get is governed by two orthogonal axes.

Scope — where the searchable content lives#

• Entity — the field is indexed on its own row, for entity-local field search. You query it with the field primitives Prop.Matches(q), Prop.TextScore(q), and Prop.Similarity(q) and compose your own ranking (see field search (Matches / TextScore / Similarity)). This is the default for a String property.
• Memory — the field's text is chunked into the app's shared corpus, the cross-entity store that using Memory (semantic search) searches with one turnkey call. This is the default for a Markdown property (Markdown is typically long and sectioned, so the corpus is its natural home).

If you don't write a scope, it's chosen by the property's type: String → Entity, Markdown → Memory. Write the scope explicitly to override — e.g. [Searchable(Memory)] string Summary; puts a short String field into the corpus, and [Searchable(Entity)] Markdown Body; keeps a Markdown field's search entity-local.

Mode — which kinds of relevance#

• Full (default) — both lexical (full-text keyword match) and semantic (meaning-based, vector) ranking. Semantic ranking is active whenever an embedding provider is configured; without one, the field still works as full-text and upgrades automatically once an embedder is wired.
• TextOnly — lexical only. No embedder is ever needed, and the field carries no per-row vector. Use it when keyword search is all you want and you don't want the storage or the dependency. TextOnly applies to Entity scope only — the shared corpus is always hybrid, so [Searchable(Memory, TextOnly)] is rejected.

Degrade, don't fail#

Full-text needs no external service, so a [Searchable] field is useful the instant you deploy. Semantic ranking is an upgrade: declare an embedding for the app and every Full / Memory field starts ranking by meaning too — no code change. If you deploy Full or Memory searchable fields with no embedder configured, the deploy succeeds and warns that semantic ranking is inactive until you wire one; full-text is live in the meantime.

Examples#

using Osyrin.Memory;

entity Article {
  [MaxLength(200)] string Title;
  [Searchable] string Body;            // Entity scope, Full mode — the String default
}

using Osyrin.Memory;

entity Note {
  [Searchable(Entity, TextOnly)] string Text;   // full-text keyword search only
}

using Osyrin.Memory;

entity Doc {
  [Searchable(Memory)] string Summary;   // a String explicitly placed in the corpus
  [Searchable] Markdown Body;            // Markdown defaults to the corpus
}

See also#

• field search (Matches / TextScore / Similarity) — the field primitives (Matches/TextScore/Similarity) for Entity-scope fields
• using Memory (semantic search) — the turnkey corpus search over Memory-scope fields
• SearchHit — the result type Memory.Search returns