@hollware/core 0.3.0 → 0.5.0

package/dist/index.d.mts

@@ -290,6 +290,13 @@ interface SiteDefinition {
   readonly styleGuide?: string;
   /** The public origin of the site, such as `https://example.com`: entries' URLs start with it. */
   readonly url?: string;
+  /**
+   * The origins whose pages may read the published content of the site from the browser, over
+   * the read API with a read token (`hw_read_`). None by default.
+   */
+  readonly cors?: {
+    readonly origins: readonly string[];
+  };
   /** Who hears about publications (ADR-025). */
   readonly events?: {
     /** Each gets every event, signed with HMAC-SHA256 by the Worker secret named in `secret`. */
@@ -346,6 +353,10 @@ interface ResolvedSite {
   readonly collections: ReadonlyMap<string, ResolvedCollection>;
   /** The public origin of the site, or `null`. */
   readonly url: string | null;
+  /** The origins that may call the read API from a page, each as `URL.origin` gives it. */
+  readonly cors: {
+    readonly origins: readonly string[];
+  };
   readonly events: ResolvedEvents;
 }
 /** A webhook of a site; `secret` names the Worker secret that signs its deliveries. */
@@ -361,7 +372,7 @@ interface ResolvedEvents {
     readonly key: string;
   } | null;
 }
-/** The config once resolved, at the start of the Worker. v0.1 holds exactly one site. */
+/** The config once resolved, at the start of the Worker: one site or more (ADR-027). */
 interface ResolvedConfig {
   readonly sites: ReadonlyMap<string, ResolvedSite>;
   readonly auth: ResolvedAuth;
@@ -428,7 +439,7 @@ export declare const DEFAULT_MAX_MARKDOWN_BYTES = 524288;
  */
 export declare const MAX_MARKDOWN_BYTES_CEILING = 786432;
 declare namespace schemas_d_exports {
-  export { acknowledgedWarning, collectionDescription, collectionName, componentDescription, conflict, contractDescription, document, entry, entryId, entryStatus, entrySummary, format, locale, markdown, phase, publication, review, reviewId, reviewStatus, revisionId, severity, siteDescription, slug, verdict, version, violation };
+  export { acknowledgedWarning, collectionDescription, collectionName, componentDescription, conflict, contractDescription, document, entry, entryId, entryStatus, entrySummary, format, locale, markdown, phase, publication, review, reviewId, reviewStatus, revisionId, searchMode, severity, siteDescription, slug, verdict, version, violation };
 }
 /**
  * The pieces the operation schemas share. Their descriptions become the JSON Schema that agents
@@ -692,6 +703,12 @@ declare const format: z.ZodEnum<{
   html: "html";
   markdown: "markdown";
 }>;
+/** How a search is made: by its words, by meaning, or both merged (ADR-030). */
+declare const searchMode: z.ZodEnum<{
+  fulltext: "fulltext";
+  hybrid: "hybrid";
+  semantic: "semantic";
+}>;
 declare const componentDescription: z.ZodObject<{
   name: z.ZodString;
   description: z.ZodNullable<z.ZodString>;
@@ -954,7 +971,7 @@ export declare function describeResolvedSite(site: ResolvedSite, options?: {
 /**
  * Checks the config of the deployment and fills in its defaults, or throws a `ConfigError` with
  * every problem it found, so that a broken `hollware.config.ts` stops the Worker at startup with
- * a message that says what to change. v0.1 serves exactly one site per deployment.
+ * a message that says what to change. A deployment serves one site or more.
  *
  * The parameter is `unknown` because the config comes from user code that may be plain
  * JavaScript; a `HollwareConfig` is accepted as is, so this is a `ResolveConfig`. Notes on the
@@ -1037,6 +1054,7 @@ export declare class HollwareError extends Error {
 export declare const ID_PREFIXES: {
   readonly challenge: 'chl';
   readonly delivery: 'dlv';
+  readonly embedding: 'emb';
   readonly entry: 'ent';
   readonly event: 'evt';
   readonly invitation: 'inv';
@@ -1227,6 +1245,48 @@ export declare const listEntries: OperationSpec<"list_entries", z.ZodObject<{
   }, z.core.$strip>>;
   nextCursor: z.ZodNullable<z.ZodString>;
 }, z.core.$strip>>;
+export declare const search: OperationSpec<"search", z.ZodObject<{
+  query: z.ZodString;
+  collection: z.ZodOptional<z.ZodString>;
+  locale: z.ZodOptional<z.ZodString>;
+  limit: z.ZodDefault<z.ZodNumber>;
+  mode: z.ZodOptional<z.ZodEnum<{
+    fulltext: "fulltext";
+    hybrid: "hybrid";
+    semantic: "semantic";
+  }>>;
+}, z.core.$strict>, z.ZodObject<{
+  outcome: z.ZodEnum<{
+    ok: "ok";
+  }>;
+  mode: z.ZodEnum<{
+    fulltext: "fulltext";
+    hybrid: "hybrid";
+    semantic: "semantic";
+  }>;
+  results: z.ZodArray<z.ZodObject<{
+    entry: z.ZodObject<{
+      id: z.ZodString;
+      collection: z.ZodString;
+      locale: z.ZodString;
+      slug: z.ZodString;
+      translationGroup: z.ZodNullable<z.ZodString>;
+      status: z.ZodEnum<{
+        draft: "draft";
+        published: "published";
+        scheduled: "scheduled";
+      }>;
+      version: z.ZodNumber;
+      frontmatter: z.ZodRecord<z.ZodString, z.ZodUnknown>;
+      tags: z.ZodArray<z.ZodString>;
+      hasUnpublishedChanges: z.ZodBoolean;
+      publishedAt: z.ZodNullable<z.ZodString>;
+      updatedAt: z.ZodString;
+    }, z.core.$strip>;
+    snippet: z.ZodString;
+    score: z.ZodNumber;
+  }, z.core.$strip>>;
+}, z.core.$strip>>;
 export declare const getEntry: OperationSpec<"get_entry", z.ZodObject<{
   collection: z.ZodString;
   locale: z.ZodString;
@@ -1237,6 +1297,7 @@ export declare const getEntry: OperationSpec<"get_entry", z.ZodObject<{
     markdown: "markdown";
   }>>>;
   published: z.ZodDefault<z.ZodBoolean>;
+  related: z.ZodDefault<z.ZodBoolean>;
 }, z.core.$strict>, z.ZodObject<{
   outcome: z.ZodEnum<{
     not_found: "not_found";
@@ -1336,6 +1397,24 @@ export declare const getEntry: OperationSpec<"get_entry", z.ZodObject<{
     }, z.core.$strip>>;
     html: z.ZodOptional<z.ZodString>;
   }, z.core.$strip>>;
+  related: z.ZodOptional<z.ZodArray<z.ZodObject<{
+    id: z.ZodString;
+    collection: z.ZodString;
+    locale: z.ZodString;
+    slug: z.ZodString;
+    translationGroup: z.ZodNullable<z.ZodString>;
+    status: z.ZodEnum<{
+      draft: "draft";
+      published: "published";
+      scheduled: "scheduled";
+    }>;
+    version: z.ZodNumber;
+    frontmatter: z.ZodRecord<z.ZodString, z.ZodUnknown>;
+    tags: z.ZodArray<z.ZodString>;
+    hasUnpublishedChanges: z.ZodBoolean;
+    publishedAt: z.ZodNullable<z.ZodString>;
+    updatedAt: z.ZodString;
+  }, z.core.$strip>>>;
 }, z.core.$strip>>;
 export declare const getTranslations: OperationSpec<"get_translations", z.ZodObject<{
   collection: z.ZodString;
@@ -1802,6 +1881,38 @@ export declare const requestChanges: OperationSpec<"request_changes", z.ZodObjec
     note: z.ZodNullable<z.ZodString>;
   }, z.core.$strip>>;
 }, z.core.$strip>>;
+export declare const contentStats: OperationSpec<"content_stats", z.ZodObject<{
+  days: z.ZodDefault<z.ZodNumber>;
+  limit: z.ZodDefault<z.ZodNumber>;
+}, z.core.$strict>, z.ZodObject<{
+  outcome: z.ZodEnum<{
+    ok: "ok";
+    unavailable: "unavailable";
+  }>;
+  since: z.ZodOptional<z.ZodString>;
+  pages: z.ZodOptional<z.ZodArray<z.ZodObject<{
+    path: z.ZodString;
+    views: z.ZodNumber;
+    entry: z.ZodNullable<z.ZodObject<{
+      id: z.ZodString;
+      collection: z.ZodString;
+      locale: z.ZodString;
+      slug: z.ZodString;
+      translationGroup: z.ZodNullable<z.ZodString>;
+      status: z.ZodEnum<{
+        draft: "draft";
+        published: "published";
+        scheduled: "scheduled";
+      }>;
+      version: z.ZodNumber;
+      frontmatter: z.ZodRecord<z.ZodString, z.ZodUnknown>;
+      tags: z.ZodArray<z.ZodString>;
+      hasUnpublishedChanges: z.ZodBoolean;
+      publishedAt: z.ZodNullable<z.ZodString>;
+      updatedAt: z.ZodString;
+    }, z.core.$strip>>;
+  }, z.core.$strip>>>;
+}, z.core.$strip>>;
 /** Every operation, sorted by name: the order of `tools/list`. */
 export declare const OPERATIONS: readonly [OperationSpec<"approve_review", z.ZodObject<{
   id: z.ZodString;
@@ -1882,6 +1993,37 @@ export declare const OPERATIONS: readonly [OperationSpec<"approve_review", z.Zod
     }, z.core.$strip>>;
     contractsHash: z.ZodString;
   }, z.core.$strip>>;
+}, z.core.$strip>>, OperationSpec<"content_stats", z.ZodObject<{
+  days: z.ZodDefault<z.ZodNumber>;
+  limit: z.ZodDefault<z.ZodNumber>;
+}, z.core.$strict>, z.ZodObject<{
+  outcome: z.ZodEnum<{
+    ok: "ok";
+    unavailable: "unavailable";
+  }>;
+  since: z.ZodOptional<z.ZodString>;
+  pages: z.ZodOptional<z.ZodArray<z.ZodObject<{
+    path: z.ZodString;
+    views: z.ZodNumber;
+    entry: z.ZodNullable<z.ZodObject<{
+      id: z.ZodString;
+      collection: z.ZodString;
+      locale: z.ZodString;
+      slug: z.ZodString;
+      translationGroup: z.ZodNullable<z.ZodString>;
+      status: z.ZodEnum<{
+        draft: "draft";
+        published: "published";
+        scheduled: "scheduled";
+      }>;
+      version: z.ZodNumber;
+      frontmatter: z.ZodRecord<z.ZodString, z.ZodUnknown>;
+      tags: z.ZodArray<z.ZodString>;
+      hasUnpublishedChanges: z.ZodBoolean;
+      publishedAt: z.ZodNullable<z.ZodString>;
+      updatedAt: z.ZodString;
+    }, z.core.$strip>>;
+  }, z.core.$strip>>>;
 }, z.core.$strip>>, OperationSpec<"create_draft", z.ZodObject<{
   collection: z.ZodString;
   locale: z.ZodString;
@@ -1988,6 +2130,7 @@ export declare const OPERATIONS: readonly [OperationSpec<"approve_review", z.Zod
     markdown: "markdown";
   }>>>;
   published: z.ZodDefault<z.ZodBoolean>;
+  related: z.ZodDefault<z.ZodBoolean>;
 }, z.core.$strict>, z.ZodObject<{
   outcome: z.ZodEnum<{
     not_found: "not_found";
@@ -2087,6 +2230,24 @@ export declare const OPERATIONS: readonly [OperationSpec<"approve_review", z.Zod
     }, z.core.$strip>>;
     html: z.ZodOptional<z.ZodString>;
   }, z.core.$strip>>;
+  related: z.ZodOptional<z.ZodArray<z.ZodObject<{
+    id: z.ZodString;
+    collection: z.ZodString;
+    locale: z.ZodString;
+    slug: z.ZodString;
+    translationGroup: z.ZodNullable<z.ZodString>;
+    status: z.ZodEnum<{
+      draft: "draft";
+      published: "published";
+      scheduled: "scheduled";
+    }>;
+    version: z.ZodNumber;
+    frontmatter: z.ZodRecord<z.ZodString, z.ZodUnknown>;
+    tags: z.ZodArray<z.ZodString>;
+    hasUnpublishedChanges: z.ZodBoolean;
+    publishedAt: z.ZodNullable<z.ZodString>;
+    updatedAt: z.ZodString;
+  }, z.core.$strip>>>;
 }, z.core.$strip>>, OperationSpec<"get_translations", z.ZodObject<{
   collection: z.ZodString;
   translationGroup: z.ZodString;
@@ -2312,6 +2473,47 @@ export declare const OPERATIONS: readonly [OperationSpec<"approve_review", z.Zod
     decidedAt: z.ZodNullable<z.ZodString>;
     note: z.ZodNullable<z.ZodString>;
   }, z.core.$strip>>;
+}, z.core.$strip>>, OperationSpec<"search", z.ZodObject<{
+  query: z.ZodString;
+  collection: z.ZodOptional<z.ZodString>;
+  locale: z.ZodOptional<z.ZodString>;
+  limit: z.ZodDefault<z.ZodNumber>;
+  mode: z.ZodOptional<z.ZodEnum<{
+    fulltext: "fulltext";
+    hybrid: "hybrid";
+    semantic: "semantic";
+  }>>;
+}, z.core.$strict>, z.ZodObject<{
+  outcome: z.ZodEnum<{
+    ok: "ok";
+  }>;
+  mode: z.ZodEnum<{
+    fulltext: "fulltext";
+    hybrid: "hybrid";
+    semantic: "semantic";
+  }>;
+  results: z.ZodArray<z.ZodObject<{
+    entry: z.ZodObject<{
+      id: z.ZodString;
+      collection: z.ZodString;
+      locale: z.ZodString;
+      slug: z.ZodString;
+      translationGroup: z.ZodNullable<z.ZodString>;
+      status: z.ZodEnum<{
+        draft: "draft";
+        published: "published";
+        scheduled: "scheduled";
+      }>;
+      version: z.ZodNumber;
+      frontmatter: z.ZodRecord<z.ZodString, z.ZodUnknown>;
+      tags: z.ZodArray<z.ZodString>;
+      hasUnpublishedChanges: z.ZodBoolean;
+      publishedAt: z.ZodNullable<z.ZodString>;
+      updatedAt: z.ZodString;
+    }, z.core.$strip>;
+    snippet: z.ZodString;
+    score: z.ZodNumber;
+  }, z.core.$strip>>;
 }, z.core.$strip>>, OperationSpec<"unpublish", z.ZodObject<{
   id: z.ZodString;
 }, z.core.$strict>, z.ZodObject<{
@@ -2456,6 +2658,11 @@ type OperationByName<Name extends OperationName> = Extract<Operation, {
  * row of a token has an id of its own (`tok_…`, see `ids.ts`), its scopes and its expiry.
  */
 export declare const PERSONAL_TOKEN_PREFIX = "hw_pat_";
+/**
+ * Read tokens: `hw_read_` and the same 52 characters. A read token reads the published content of
+ * one site over the read API (`GET /v1/…`) and nothing else, so a site may ship one to a browser.
+ */
+export declare const READ_TOKEN_PREFIX = "hw_read_";
 /** Days until a new personal token expires, unless whoever creates it asks for another expiry. */
 export declare const PERSONAL_TOKEN_TTL_DAYS = 90;
 /**
@@ -2464,8 +2671,12 @@ export declare const PERSONAL_TOKEN_TTL_DAYS = 90;
  * tests pass their own bytes.
  */
 export declare function generatePersonalToken(random?: Uint8Array): string;
+/** A new read token, made as a personal token is. */
+export declare function generateReadToken(random?: Uint8Array): string;
 /** Whether a string has the shape of a personal token. It says nothing about whether one exists. */
 export declare function isPersonalToken(value: string): boolean;
+/** Whether a string has the shape of a read token. It says nothing about whether one exists. */
+export declare function isReadToken(value: string): boolean;
 /** The SHA-256 of a token, in lowercase hex: what the `tokens` table stores and looks up. */
 export declare function hashPersonalToken(token: string): Promise<string>;
 //#endregion
@@ -2496,7 +2707,38 @@ export interface WebhookCheck {
  */
 export declare function verifyWebhook(check: WebhookCheck): Promise<boolean>;
 //#endregion
+//#region src/search/chunks.d.ts
+/**
+ * The pieces of an entry that semantic search embeds (v0.5): one summary, and one piece for each
+ * section the body has under an `h2`. Vectors are derived data, like the Comark document: a change
+ * of this version makes every vector one to make again.
+ */
+export declare const CHUNKER_VERSION = "1";
+interface Chunk {
+  /** The place of the piece in the entry: 0 is the summary, then the sections in order. */
+  readonly index: number;
+  readonly kind: 'summary' | 'section';
+  /** The heading of a section; `null` for the summary. */
+  readonly heading: string | null;
+  readonly text: string;
+}
+/** How long a piece may grow, in characters, before it is cut at a word. */
+export declare const CHUNK_LIMITS: {
+  readonly summary: 1200;
+  readonly section: 2000;
+};
+/**
+ * The pieces of an entry: the summary (its title, its description and the text before the first
+ * `h2`), then each section (its heading and its text up to the next `h2`). An `h1` adds nothing:
+ * it repeats the title. A section with no text is left out.
+ */
+export declare function chunksOf(entry: {
+  readonly title: string;
+  readonly description?: string;
+  readonly document: EntryDocument;
+}): Chunk[];
+//#endregion
 //#region src/index.d.ts
 export declare const name = "@hollware/core";
 //#endregion
-export { type AcknowledgedWarning, type AuthDefinition, type BuiltInContractId, type BuiltInOptions, type CollectionDefinition, type ComponentDefinition, type ComponentUse, type ConfigProblem, type Contract, type ContractContext, type ContractDescription, type ContractFinding, type ContractText, type CustomContractOptions, type DescribedSchema, type DisallowedMarker, type Effect, type EntryDocument, type EntryFacts, type EntryLookup, type EntryRef, type ErrorCode, type ExtractedContent, type Heading, type HollwareConfig, type IdPrefix, type ImageRef, type LinkRef, type LinkTarget, type Operation, type OperationByName, type OperationName, type OperationSpec, type OutcomeInfo, type OutcomeOf, type Phase, type ProcessMarkdown, type ProcessMarkdownOptions, type ProcessResult, type ProcessedMarkdown, type RenderHtml, type ResolveConfig, type ResolvedAuth, type ResolvedCollection, type ResolvedConfig, type ResolvedSite, type RunContractsInput, type SchemaOutput, type Scope, type Severity, type SiteDefinition, type SiteDescription, type Verdict, type Violation, schemas_d_exports as schemas };
+export { type AcknowledgedWarning, type AuthDefinition, type BuiltInContractId, type BuiltInOptions, type Chunk, type CollectionDefinition, type ComponentDefinition, type ComponentUse, type ConfigProblem, type Contract, type ContractContext, type ContractDescription, type ContractFinding, type ContractText, type CustomContractOptions, type DescribedSchema, type DisallowedMarker, type Effect, type EntryDocument, type EntryFacts, type EntryLookup, type EntryRef, type ErrorCode, type ExtractedContent, type Heading, type HollwareConfig, type IdPrefix, type ImageRef, type LinkRef, type LinkTarget, type Operation, type OperationByName, type OperationName, type OperationSpec, type OutcomeInfo, type OutcomeOf, type Phase, type ProcessMarkdown, type ProcessMarkdownOptions, type ProcessResult, type ProcessedMarkdown, type RenderHtml, type ResolveConfig, type ResolvedAuth, type ResolvedCollection, type ResolvedConfig, type ResolvedSite, type RunContractsInput, type SchemaOutput, type Scope, type Severity, type SiteDefinition, type SiteDescription, type Verdict, type Violation, schemas_d_exports as schemas };

package/dist/index.mjs

@@ -1396,6 +1396,20 @@ function readSiteUrl(path, raw, problems) {
 	}
 	return url.origin;
 }
+/** The origins of a site's pages that may read its published content from the browser. */
+function readCors(path, raw, problems) {
+	if (raw === void 0) return { origins: [] };
+	if (!isRecord$1(raw) || !Array.isArray(raw["origins"])) {
+		problems.add(path, wrongType(raw, "an object with origins"), "Declare it as { origins: [\"https://example.com\"] }, or remove it.");
+		return { origins: [] };
+	}
+	const origins = [];
+	for (const [index, origin] of raw["origins"].entries()) {
+		const read = readSiteUrl(itemPath(keyPath(path, "origins"), index), origin, problems);
+		if (read !== null && !origins.includes(read)) origins.push(read);
+	}
+	return { origins };
+}
 function readBindingName(path, raw, problems) {
 	if (typeof raw === "string" && BINDING_NAME.test(raw)) return raw;
 	problems.add(path, wrongType(raw, "the name of a Worker secret"), "Name the secret that holds the value, such as \"WEBHOOK_SECRET\", and set it with wrangler secret put.");
@@ -1573,6 +1587,7 @@ function readSite(key, raw, problems) {
 		styleGuide,
 		collections,
 		url,
+		cors: readCors(keyPath(path, "cors"), raw["cors"], problems),
 		events: readEvents(keyPath(path, "events"), raw["events"], url, problems)
 	};
 }
@@ -1583,8 +1598,7 @@ function readSites(raw, problems) {
 		return sites;
 	}
 	const keys = Object.keys(raw);
-	if (keys.length === 0) problems.add("sites", "has no site, and v0.1 serves exactly one site per deployment", SITE_FIX);
-	else if (keys.length > 1) problems.add("sites", `has ${keys.length} sites (${keys.map(quote).join(", ")}), and v0.1 serves exactly one site per deployment`, "Keep exactly one site in sites.");
+	if (keys.length === 0) problems.add("sites", "has no site, and a deployment serves at least one", SITE_FIX);
 	for (const key of keys) {
 		const site = readSite(key, raw[key], problems);
 		if (site !== null) sites.set(key, site);
@@ -1609,7 +1623,7 @@ function readAuth(raw, problems) {
 /**
 * Checks the config of the deployment and fills in its defaults, or throws a `ConfigError` with
 * every problem it found, so that a broken `hollware.config.ts` stops the Worker at startup with
-* a message that says what to change. v0.1 serves exactly one site per deployment.
+* a message that says what to change. A deployment serves one site or more.
 *
 * The parameter is `unknown` because the config comes from user code that may be plain
 * JavaScript; a `HollwareConfig` is accepted as is, so this is a `ResolveConfig`. Notes on the
@@ -1688,6 +1702,7 @@ const CROCKFORD = "0123456789ABCDEFGHJKMNPQRSTVWXYZ";
 const ID_PREFIXES = {
 	challenge: "chl",
 	delivery: "dlv",
+	embedding: "emb",
 	entry: "ent",
 	event: "evt",
 	invitation: "inv",
@@ -3038,6 +3053,7 @@ var schemas_exports = /* @__PURE__ */ __exportAll({
 	reviewId: () => reviewId,
 	reviewStatus: () => reviewStatus,
 	revisionId: () => revisionId,
+	searchMode: () => searchMode,
 	severity: () => severity,
 	siteDescription: () => siteDescription,
 	slug: () => slug,
@@ -3148,6 +3164,12 @@ const format = z.enum([
 	"document",
 	"html"
 ]);
+/** How a search is made: by its words, by meaning, or both merged (ADR-030). */
+const searchMode = z.enum([
+	"fulltext",
+	"semantic",
+	"hybrid"
+]);
 const componentDescription = z.object({
 	name: z.string(),
 	description: z.string().nullable(),
@@ -3226,6 +3248,30 @@ const listEntries = defineOperation({
 	effect: "read",
 	outcomes: { ok: { isError: false } }
 });
+const search = defineOperation({
+	name: "search",
+	title: "Search entries",
+	description: "Searches the published entries of the site, best match first. Full text finds the entries that hold every word of query, in their title, tags and text; semantic finds them by meaning, in any language; hybrid merges both. Narrow it to a collection or a locale. Drafts are not searched: read them with list_entries.",
+	input: z.strictObject({
+		query: z.string().trim().min(1).max(200).describe("The words to find, or a question; punctuation is ignored by full text."),
+		collection: collectionName.optional(),
+		locale: locale.optional(),
+		limit: z.number().int().min(1).max(50).default(10),
+		mode: searchMode.optional().describe("fulltext, semantic or hybrid; hybrid when not given. Semantic and hybrid need the semantic search of the deployment: without it the search is full text, and mode in the answer says so.")
+	}),
+	output: z.object({
+		outcome: z.enum(["ok"]),
+		mode: searchMode.describe("How the search was made."),
+		results: z.array(z.object({
+			entry: entrySummary,
+			snippet: z.string().describe("The words around the match, the words found between [ and ]; for a match by meaning, the start of the closest part."),
+			score: z.number().describe("How well it matches: higher is better.")
+		}))
+	}),
+	scopes: ["content:read"],
+	effect: "read",
+	outcomes: { ok: { isError: false } }
+});
 const getEntry = defineOperation({
 	name: "get_entry",
 	title: "Get an entry",
@@ -3235,11 +3281,13 @@ const getEntry = defineOperation({
 		locale,
 		slug,
 		formats: z.array(format).min(1).default(["markdown"]).describe("What to return of the revision: markdown, the Comark document, or html."),
-		published: z.boolean().default(false).describe("Read the revision that is live instead of the latest one: not_found when the entry is not published.")
+		published: z.boolean().default(false).describe("Read the revision that is live instead of the latest one: not_found when the entry is not published."),
+		related: z.boolean().default(false).describe("Also return the published entries closest to this one by meaning, in its collection and locale, closest first. Needs the semantic search of the deployment; empty without it.")
 	}),
 	output: z.object({
 		outcome: z.enum(["ok", "not_found"]),
-		entry: entry.optional()
+		entry: entry.optional(),
+		related: z.array(entrySummary).optional().describe("When related was asked for.")
 	}),
 	scopes: ["content:read"],
 	effect: "read",
@@ -3466,9 +3514,34 @@ const requestChanges = defineOperation({
 		not_found: { isError: true }
 	}
 });
+const contentStats = defineOperation({
+	name: "content_stats",
+	title: "Read what readers read",
+	description: "The most read pages of the sites that read this one, over the last days, most read first, each with the published entry it shows when it shows one. Use it to pick what to update. unavailable when the deployment has no analytics.",
+	input: z.strictObject({
+		days: z.number().int().min(1).max(90).default(30).describe("How many days back to count."),
+		limit: z.number().int().min(1).max(50).default(10)
+	}),
+	output: z.object({
+		outcome: z.enum(["ok", "unavailable"]),
+		since: z.string().optional().describe("The start of the count, in ISO 8601."),
+		pages: z.array(z.object({
+			path: z.string(),
+			views: z.number().describe("Page views, sampling accounted for."),
+			entry: entrySummary.nullable().describe("The published entry at that path, or null.")
+		})).optional()
+	}),
+	scopes: ["content:read"],
+	effect: "read",
+	outcomes: {
+		ok: { isError: false },
+		unavailable: { isError: true }
+	}
+});
 /** Every operation, sorted by name: the order of `tools/list`. */
 const OPERATIONS = [
 	approveReview,
+	contentStats,
 	createDraft,
 	describeSite,
 	getEntry,
@@ -3477,6 +3550,7 @@ const OPERATIONS = [
 	listReviews,
 	publish,
 	requestChanges,
+	search,
 	unpublish,
 	updateDraft,
 	validate
@@ -3506,26 +3580,43 @@ function isScope(value) {
 * row of a token has an id of its own (`tok_…`, see `ids.ts`), its scopes and its expiry.
 */
 const PERSONAL_TOKEN_PREFIX = "hw_pat_";
+/**
+* Read tokens: `hw_read_` and the same 52 characters. A read token reads the published content of
+* one site over the read API (`GET /v1/…`) and nothing else, so a site may ship one to a browser.
+*/
+const READ_TOKEN_PREFIX = "hw_read_";
 /** Days until a new personal token expires, unless whoever creates it asks for another expiry. */
 const PERSONAL_TOKEN_TTL_DAYS = 90;
 const ALPHABET = "0123456789abcdefghjkmnpqrstvwxyz";
 const SECRET_LENGTH = 52;
 const TOKEN_PATTERN = /^hw_pat_[0-9a-hjkmnp-tv-z]{52}$/;
+const READ_TOKEN_PATTERN = /^hw_read_[0-9a-hjkmnp-tv-z]{52}$/;
 /**
 * A new personal token. Each of the 52 random bytes gives one character (its low 5 bits, which
 * are uniform because 256 is a multiple of 32). `random` defaults to `crypto.getRandomValues`;
 * tests pass their own bytes.
 */
 function generatePersonalToken(random = crypto.getRandomValues(new Uint8Array(SECRET_LENGTH))) {
-	if (random.length !== SECRET_LENGTH) throw new RangeError(`a personal token takes ${SECRET_LENGTH} random bytes, got ${random.length}`);
+	return PERSONAL_TOKEN_PREFIX + secretOf(random, "a personal token");
+}
+/** A new read token, made as a personal token is. */
+function generateReadToken(random = crypto.getRandomValues(new Uint8Array(SECRET_LENGTH))) {
+	return READ_TOKEN_PREFIX + secretOf(random, "a read token");
+}
+function secretOf(random, what) {
+	if (random.length !== SECRET_LENGTH) throw new RangeError(`${what} takes ${SECRET_LENGTH} random bytes, got ${random.length}`);
 	let secret = "";
 	for (const byte of random) secret += ALPHABET.charAt(byte & 31);
-	return PERSONAL_TOKEN_PREFIX + secret;
+	return secret;
 }
 /** Whether a string has the shape of a personal token. It says nothing about whether one exists. */
 function isPersonalToken(value) {
 	return TOKEN_PATTERN.test(value);
 }
+/** Whether a string has the shape of a read token. It says nothing about whether one exists. */
+function isReadToken(value) {
+	return READ_TOKEN_PATTERN.test(value);
+}
 /** The SHA-256 of a token, in lowercase hex: what the `tokens` table stores and looks up. */
 async function hashPersonalToken(token) {
 	const digest = await crypto.subtle.digest("SHA-256", new TextEncoder().encode(token));
@@ -3576,7 +3667,82 @@ async function verifyWebhook(check) {
 	return sameText(signature, await signWebhook(check.secret, timestamp, check.body));
 }
 //#endregion
+//#region src/search/chunks.ts
+/**
+* The pieces of an entry that semantic search embeds (v0.5): one summary, and one piece for each
+* section the body has under an `h2`. Vectors are derived data, like the Comark document: a change
+* of this version makes every vector one to make again.
+*/
+const CHUNKER_VERSION = "1";
+/** How long a piece may grow, in characters, before it is cut at a word. */
+const CHUNK_LIMITS = {
+	summary: 1200,
+	section: 2e3
+};
+/** The text a reader sees in a node: its strings, never its tag, attributes or comments. */
+function textOf(node) {
+	if (typeof node === "string") return node;
+	if (!Array.isArray(node) || node[0] === null) return "";
+	return node.slice(2).map((child) => textOf(child)).join("");
+}
+function tagOf(node) {
+	return Array.isArray(node) && typeof node[0] === "string" ? node[0] : null;
+}
+/** The words of a text in one line, cut at the last word that fits in `limit`. */
+function fit(text, limit) {
+	const flat = text.replace(/\s+/g, " ").trim();
+	if (flat.length <= limit) return flat;
+	const cut = flat.slice(0, limit);
+	const lastSpace = cut.lastIndexOf(" ");
+	return (lastSpace > 0 ? cut.slice(0, lastSpace) : cut).trim();
+}
+/**
+* The pieces of an entry: the summary (its title, its description and the text before the first
+* `h2`), then each section (its heading and its text up to the next `h2`). An `h1` adds nothing:
+* it repeats the title. A section with no text is left out.
+*/
+function chunksOf(entry) {
+	const intro = [];
+	const sections = [];
+	for (const node of entry.document.nodes) {
+		const tag = tagOf(node);
+		if (tag === "h1") continue;
+		if (tag === "h2") {
+			sections.push({
+				heading: textOf(node).trim(),
+				parts: []
+			});
+			continue;
+		}
+		const text = textOf(node);
+		const current = sections.at(-1);
+		if (current === void 0) intro.push(text);
+		else current.parts.push(text);
+	}
+	const chunks = [{
+		index: 0,
+		kind: "summary",
+		heading: null,
+		text: fit([
+			entry.title,
+			entry.description ?? "",
+			intro.join(" ")
+		].filter((part) => part !== "").join("\n"), CHUNK_LIMITS.summary)
+	}];
+	for (const section of sections) {
+		const body = section.parts.join(" ").trim();
+		if (body === "") continue;
+		chunks.push({
+			index: chunks.length,
+			kind: "section",
+			heading: section.heading,
+			text: fit(`${section.heading}\n${body}`, CHUNK_LIMITS.section)
+		});
+	}
+	return chunks;
+}
+//#endregion
 //#region src/index.ts
 const name = "@hollware/core";
 //#endregion
-export { BUILT_IN_CONTRACT_IDS, COMARK_VERSION, ConfigError, DEFAULT_MAX_MARKDOWN_BYTES, ERROR_CODES, EVENT_ID_HEADER, HollwareError, ID_PREFIXES, MAX_MARKDOWN_BYTES_CEILING, OPERATIONS, PERSONAL_TOKEN_PREFIX, PERSONAL_TOKEN_TTL_DAYS, PIPELINE_VERSION, REFUSAL_CONTRACT_IDS, SCOPES, SIGNATURE_HEADER, SITE_KEY_MAX_LENGTH, SITE_KEY_PATTERN, TIMESTAMP_HEADER, approveReview, contracts, createDraft, defineCollection, defineComponent, defineConfig, defineContract, defineOperation, describeBuiltInContracts, describeContract, describeResolvedSite, describeSite, generatePersonalToken, getEntry, getTranslations, hashPersonalToken, isBuiltInContract, isId, isPersonalToken, isScope, isSiteKey, listEntries, listReviews, name, newId, processMarkdown, publish, renderHtml, requestChanges, resolveConfig, runContracts, schemas_exports as schemas, signWebhook, sizeLimitViolation, ulid, uniqueSlugViolation, uniqueTranslationViolation, unpublish, updateDraft, validate, verifyWebhook };
+export { BUILT_IN_CONTRACT_IDS, CHUNKER_VERSION, CHUNK_LIMITS, COMARK_VERSION, ConfigError, DEFAULT_MAX_MARKDOWN_BYTES, ERROR_CODES, EVENT_ID_HEADER, HollwareError, ID_PREFIXES, MAX_MARKDOWN_BYTES_CEILING, OPERATIONS, PERSONAL_TOKEN_PREFIX, PERSONAL_TOKEN_TTL_DAYS, PIPELINE_VERSION, READ_TOKEN_PREFIX, REFUSAL_CONTRACT_IDS, SCOPES, SIGNATURE_HEADER, SITE_KEY_MAX_LENGTH, SITE_KEY_PATTERN, TIMESTAMP_HEADER, approveReview, chunksOf, contentStats, contracts, createDraft, defineCollection, defineComponent, defineConfig, defineContract, defineOperation, describeBuiltInContracts, describeContract, describeResolvedSite, describeSite, generatePersonalToken, generateReadToken, getEntry, getTranslations, hashPersonalToken, isBuiltInContract, isId, isPersonalToken, isReadToken, isScope, isSiteKey, listEntries, listReviews, name, newId, processMarkdown, publish, renderHtml, requestChanges, resolveConfig, runContracts, schemas_exports as schemas, search, signWebhook, sizeLimitViolation, ulid, uniqueSlugViolation, uniqueTranslationViolation, unpublish, updateDraft, validate, verifyWebhook };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hollware/core",
-  "version": "0.3.0",
+  "version": "0.5.0",
   "description": "Contracts, operation specs, schemas and the Markdown pipeline of Hollware. No Cloudflare dependencies.",
   "homepage": "https://hollware.com",
   "bugs": {