@lunora/ai 1.0.0-alpha.17 → 1.0.0-alpha.19

package/dist/packem_shared/{defineRag-CH8wZ0d5.mjs → defineRag-DgVXmNGw.mjs}

@@ -109,6 +109,10 @@ const resolveEmbeddingModel = (input, ai) => {
   }
   return ai.embeddingModel(input);
 };
+const modelIdOf = (model) => {
+  const id = model.modelId;
+  return typeof id === "string" && id.length > 0 ? id : void 0;
+};
 const defineRag = (config) => {
   if (typeof config.index !== "string" || config.index.length === 0) {
     throw new LunoraError("BAD_REQUEST", "@lunora/ai/rag: `index` must be a non-empty Vectorize index name");
@@ -143,10 +147,22 @@ const defineRag = (config) => {
   };
   return (context) => {
     let model;
+    const tracer = typeof context.trace === "function" ? context.trace : void 0;
     const embedText = async (text) => {
       model ??= resolveEmbeddingModel(config.embeddingModel, context.ai);
-      const { embedding } = await embed({ model, value: text });
-      return embedding;
+      const resolvedModel = model;
+      const run = async () => {
+        const { embedding } = await embed({ model: resolvedModel, value: text });
+        return embedding;
+      };
+      if (tracer === void 0) {
+        return run();
+      }
+      const modelId = modelIdOf(resolvedModel);
+      return tracer("ai.embed", run, {
+        "gen_ai.operation.name": "embeddings",
+        ...modelId === void 0 ? {} : { "gen_ai.request.model": modelId }
+      });
     };
     const checkNamespace = (namespace) => {
       if (namespace !== void 0) {

package/dist/rag/index.d.mts

@@ -109,6 +109,16 @@ interface RagContext {
    * stays decoupled from `@lunora/server`'s identity type; `rlsFilter` narrows.
    */
   auth?: unknown;
+  /**
+   * Optional `ctx.trace` span factory — an `ActionCtx`'s `ctx.trace` satisfies
+   * it structurally. When present, `defineRag` wraps each embedding-model
+   * call in a `generation` span (`gen_ai.operation.name: "embeddings"` +
+   * `gen_ai.request.model`), so the embed shows up on the trace waterfall like
+   * any other instrumented sub-operation. `unknown` on purpose — the same
+   * decoupling rationale as `auth`: `defineRag` narrows it to a callable and
+   * runs embeds untraced when it is absent (a hand-built context / test).
+   */
+  trace?: unknown;
   vectors: RagVectors;
 }
 /**
@@ -476,30 +486,6 @@ interface Rag {
   /** Embed the query and return ranked chunks + prompt-ready context. */
   retrieve: (query: string, options?: RetrieveOptions) => Promise<RetrieveResult>;
 }
-/**
- * Declare a RAG index over the two facades every Lunora action already has:
- * `ctx.ai` (embeddings) + `ctx.vectors` (Vectorize). Returns a per-request
- * factory — bind a ctx to get `{ index, retrieve, remove, asTool }`:
- *
- * ```ts
- * // lunora/rag.ts
- * export const docs = defineRag({ embeddingModel: "@cf/baai/bge-base-en-v1.5", index: "docs" });
- *
- * // inside an action:
- * await docs(ctx).index({ id: doc._id, metadata: { title: doc.title }, namespace: ctx.shardKey, text: doc.body });
- * const { chunks, context, sources } = await docs(ctx).retrieve(question, { namespace: ctx.shardKey });
- * ```
- *
- * Pure composition — no new binding, no I/O until a method runs. Chunk text
- * lives in vector metadata by default (`topK` ≤ 20); supply `textStore` to move
- * it into your own storage and lift the ceiling to 100 — see `RagTextStore`.
- *
- * `ctx.ai` is only needed to resolve a Workers AI `embeddingModel` id. Pass a
- * direct AI SDK `EmbeddingModel` object (`@ai-sdk/openai`, …) and the helper
- * embeds through it without `ctx.ai` — so a bring-your-own-embeddings index
- * needs no `env.AI` binding (bind any context carrying just `vectors`).
- * @experimental
- */
 declare const defineRag: (config: RagConfig) => ((context: RagContext) => Rag);
 /**
  * Guess a MIME type from a file extension. Lowercases and strips a leading `.`

package/dist/rag/index.d.ts

@@ -109,6 +109,16 @@ interface RagContext {
    * stays decoupled from `@lunora/server`'s identity type; `rlsFilter` narrows.
    */
   auth?: unknown;
+  /**
+   * Optional `ctx.trace` span factory — an `ActionCtx`'s `ctx.trace` satisfies
+   * it structurally. When present, `defineRag` wraps each embedding-model
+   * call in a `generation` span (`gen_ai.operation.name: "embeddings"` +
+   * `gen_ai.request.model`), so the embed shows up on the trace waterfall like
+   * any other instrumented sub-operation. `unknown` on purpose — the same
+   * decoupling rationale as `auth`: `defineRag` narrows it to a callable and
+   * runs embeds untraced when it is absent (a hand-built context / test).
+   */
+  trace?: unknown;
   vectors: RagVectors;
 }
 /**
@@ -476,30 +486,6 @@ interface Rag {
   /** Embed the query and return ranked chunks + prompt-ready context. */
   retrieve: (query: string, options?: RetrieveOptions) => Promise<RetrieveResult>;
 }
-/**
- * Declare a RAG index over the two facades every Lunora action already has:
- * `ctx.ai` (embeddings) + `ctx.vectors` (Vectorize). Returns a per-request
- * factory — bind a ctx to get `{ index, retrieve, remove, asTool }`:
- *
- * ```ts
- * // lunora/rag.ts
- * export const docs = defineRag({ embeddingModel: "@cf/baai/bge-base-en-v1.5", index: "docs" });
- *
- * // inside an action:
- * await docs(ctx).index({ id: doc._id, metadata: { title: doc.title }, namespace: ctx.shardKey, text: doc.body });
- * const { chunks, context, sources } = await docs(ctx).retrieve(question, { namespace: ctx.shardKey });
- * ```
- *
- * Pure composition — no new binding, no I/O until a method runs. Chunk text
- * lives in vector metadata by default (`topK` ≤ 20); supply `textStore` to move
- * it into your own storage and lift the ceiling to 100 — see `RagTextStore`.
- *
- * `ctx.ai` is only needed to resolve a Workers AI `embeddingModel` id. Pass a
- * direct AI SDK `EmbeddingModel` object (`@ai-sdk/openai`, …) and the helper
- * embeds through it without `ctx.ai` — so a bring-your-own-embeddings index
- * needs no `env.AI` binding (bind any context carrying just `vectors`).
- * @experimental
- */
 declare const defineRag: (config: RagConfig) => ((context: RagContext) => Rag);
 /**
  * Guess a MIME type from a file extension. Lowercases and strips a leading `.`

package/dist/rag/index.mjs

@@ -1,5 +1,5 @@
 export { default as fixedWindowChunks } from '../packem_shared/fixedWindowChunks-J-WfQDw9.mjs';
-export { default as defineRag } from '../packem_shared/defineRag-CH8wZ0d5.mjs';
+export { default as defineRag } from '../packem_shared/defineRag-DgVXmNGw.mjs';
 export { contentHash, guessMimeTypeFromExtension } from '../packem_shared/contentHash-Cgz5KRGD.mjs';
 export { default as hybridRank } from '../packem_shared/hybridRank-DPC9c2ON.mjs';
 export { default as bm25LexicalStore } from '../packem_shared/bm25LexicalStore-BQzWLMqX.mjs';

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lunora/ai",
-  "version": "1.0.0-alpha.17",
+  "version": "1.0.0-alpha.19",
   "description": "Workers AI helper for Lunora: provider-agnostic AI SDK access from functions, Workers AI by default",
   "keywords": [
     "ai",