@effect-uai/jina 0.9.0 → 0.10.0

package/dist/JinaReader.d.mts

@@ -0,0 +1,50 @@
+import { JinaEngine } from "./models.mjs";
+import { Context, Effect, Layer, Redacted } from "effect";
+import { HttpClient } from "effect/unstable/http";
+import * as AiError from "@effect-uai/core/AiError";
+import { CommonReadRequest, ReadResponse, WebRead } from "@effect-uai/core/WebRead";
+
+//#region src/JinaReader.d.ts
+declare namespace JinaReader_d_exports {
+  export { Config, JinaReadRequest, JinaReader, JinaReaderService, layer, make };
+}
+/**
+ * Jina-typed read request. Widens {@link CommonReadRequest} with Jina
+ * Reader's own header knobs; the common fields map onto Reader headers (see
+ * the codec below).
+ */
+type JinaReadRequest = CommonReadRequest & {
+  /** Fetching engine - see {@link JinaEngine}. */readonly engine?: JinaEngine; /** CSS selector to extract instead of the whole page (`X-Target-Selector`). */
+  readonly targetSelector?: string; /** Bypass Jina's page cache (`X-No-Cache`). */
+  readonly noCache?: boolean; /** Append a deduplicated link list, populating {@link ReadResponse.links}. */
+  readonly withLinksSummary?: boolean;
+};
+type JinaReaderService = {
+  readonly read: (request: JinaReadRequest) => Effect.Effect<ReadResponse, AiError.AiError>;
+};
+declare const JinaReader_base: Context.ServiceClass<JinaReader, "@betalyra/effect-uai/providers/jina/JinaReader", JinaReaderService>;
+/**
+ * Provider-typed service tag. Yield this for Jina-typed reads (its own
+ * Reader headers); yield the generic `WebRead` tag for provider-portable
+ * code. Both are registered by {@link layer}.
+ */
+declare class JinaReader extends JinaReader_base {}
+type Config = {
+  readonly apiKey: Redacted.Redacted;
+  readonly baseUrl?: string;
+};
+/**
+ * Build a `JinaReaderService` value. For Layer-based setup, prefer
+ * {@link layer}.
+ */
+declare const make: (cfg: Config) => Effect.Effect<JinaReaderService, never, HttpClient.HttpClient>;
+/**
+ * Layer registering both the provider-typed `JinaReader` tag and the generic
+ * `WebRead` tag over one implementation. A `CommonReadRequest` is
+ * structurally a `JinaReadRequest` with the extra knobs unset, so the generic
+ * registration forwards directly.
+ */
+declare const layer: (cfg: Config) => Layer.Layer<JinaReader | WebRead, never, HttpClient.HttpClient>;
+//#endregion
+export { Config, JinaReadRequest, JinaReader, JinaReaderService, layer, make, JinaReader_d_exports as t };
+//# sourceMappingURL=JinaReader.d.mts.map

package/dist/JinaReader.d.mts.map

@@ -0,0 +1 @@
+{"version":3,"file":"JinaReader.d.mts","names":[],"sources":["../src/JinaReader.ts"],"mappings":";;;;;;;;;;;;;;;KAoBY,eAAA,GAAkB,iBAAA;2DAEnB,MAAA,GAAS,UAAA;WAET,cAAA;WAEA,OAAA,YANgB;EAAA,SAQhB,gBAAA;AAAA;AAAA,KAGC,iBAAA;EAAA,SACD,IAAA,GAAO,OAAA,EAAS,eAAA,KAAoB,MAAA,CAAO,MAAA,CAAO,YAAA,EAAc,OAAA,CAAQ,OAAA;AAAA;AAAA,cAClF,eAAA;;;;;AAFD;cASa,UAAA,SAAmB,eAAA;AAAA,KAIpB,MAAA;EAAA,SACD,MAAA,EAAQ,QAAA,CAAS,QAAA;EAAA,SACjB,OAAA;AAAA;;;;;cAkIE,IAAA,GAAQ,GAAA,EAAK,MAAA,KAAS,MAAA,CAAO,MAAA,CAAO,iBAAA,SAA0B,UAAA,CAAW,UAAA;;;;;;;cAYzE,KAAA,GACX,GAAA,EAAK,MAAA,KACJ,KAAA,CAAM,KAAA,CAAM,UAAA,GAAa,OAAA,SAAgB,UAAA,CAAW,UAAA"}

package/dist/JinaReader.mjs

@@ -0,0 +1,130 @@
+import { t as __exportAll } from "./chunk-CfYAbeIz.mjs";
+import { Context, Duration, Effect, Layer, Schema } from "effect";
+import { HttpClient, HttpClientRequest } from "effect/unstable/http";
+import * as AiError from "@effect-uai/core/AiError";
+import { WebRead } from "@effect-uai/core/WebRead";
+//#region src/JinaReader.ts
+var JinaReader_exports = /* @__PURE__ */ __exportAll({
+	JinaReader: () => JinaReader,
+	layer: () => layer,
+	make: () => make
+});
+/**
+* Provider-typed service tag. Yield this for Jina-typed reads (its own
+* Reader headers); yield the generic `WebRead` tag for provider-portable
+* code. Both are registered by {@link layer}.
+*/
+var JinaReader = class extends Context.Service()("@betalyra/effect-uai/providers/jina/JinaReader") {};
+const buildHeaders = (request) => {
+	return {
+		Accept: "application/json",
+		"X-Return-Format": request.format ?? "markdown",
+		...request.timeout !== void 0 && { "X-Timeout": String(Math.max(1, Math.round(Duration.toMillis(request.timeout) / 1e3))) },
+		...request.engine !== void 0 && { "X-Engine": request.engine },
+		...request.targetSelector !== void 0 && { "X-Target-Selector": request.targetSelector },
+		...request.noCache === true && { "X-No-Cache": "true" },
+		...request.withLinksSummary === true && { "X-With-Links-Summary": "true" }
+	};
+};
+const WireData = Schema.Struct({
+	title: Schema.optional(Schema.NullOr(Schema.String)),
+	url: Schema.optional(Schema.NullOr(Schema.String)),
+	content: Schema.optional(Schema.NullOr(Schema.String)),
+	html: Schema.optional(Schema.NullOr(Schema.String)),
+	links: Schema.optional(Schema.Unknown)
+});
+const WireResponse = Schema.Struct({
+	code: Schema.optional(Schema.Number),
+	status: Schema.optional(Schema.Number),
+	data: WireData
+});
+const extractLinks = (links) => {
+	const urls = (Array.isArray(links) ? links : links !== null && typeof links === "object" ? Object.values(links) : []).filter((v) => typeof v === "string");
+	return urls.length > 0 ? urls : void 0;
+};
+const toResponse = (wire, request) => {
+	const { data } = wire;
+	const content = ((request.format ?? "markdown") === "html" ? data.html : data.content) ?? "";
+	const title = data.title ?? void 0;
+	const links = extractLinks(data.links);
+	return {
+		url: data.url ?? request.url,
+		content,
+		...title !== void 0 && { title },
+		...links !== void 0 && { links },
+		raw: wire
+	};
+};
+const transportFailure = (cause) => new AiError.Unavailable({
+	provider: "jina",
+	raw: cause
+});
+const httpStatusError = (status, body) => {
+	const provider = "jina";
+	const raw = body;
+	if (status === 429) return new AiError.RateLimited({
+		provider,
+		raw
+	});
+	if (status === 408 || status === 504) return new AiError.Timeout({
+		provider,
+		raw
+	});
+	if (status === 401) return new AiError.AuthFailed({
+		provider,
+		subtype: "auth",
+		raw
+	});
+	if (status === 403) return new AiError.AuthFailed({
+		provider,
+		subtype: "permission",
+		raw
+	});
+	if (status === 402) return new AiError.AuthFailed({
+		provider,
+		subtype: "billing",
+		raw
+	});
+	if (status >= 500) return new AiError.Unavailable({
+		provider,
+		status,
+		raw
+	});
+	return new AiError.InvalidRequest({
+		provider,
+		raw
+	});
+};
+const baseUrl = (cfg) => cfg.baseUrl ?? "https://r.jina.ai";
+const getRead = (cfg, request) => Effect.gen(function* () {
+	const client = yield* HttpClient.HttpClient;
+	const httpRequest = HttpClientRequest.get(`${baseUrl(cfg)}/${request.url}`).pipe(HttpClientRequest.bearerToken(cfg.apiKey), HttpClientRequest.setHeaders(buildHeaders(request)));
+	const response = yield* client.execute(httpRequest).pipe(Effect.mapError(transportFailure));
+	if (response.status >= 400) {
+		const text = yield* response.text.pipe(Effect.orElseSucceed(() => ""));
+		return yield* httpStatusError(response.status, text);
+	}
+	const json = yield* response.json.pipe(Effect.mapError(transportFailure));
+	return yield* Schema.decodeUnknownEffect(WireResponse)(json).pipe(Effect.mapError(transportFailure));
+});
+const readImpl = (cfg) => (request) => Effect.map(getRead(cfg, request), (wire) => toResponse(wire, request));
+/**
+* Build a `JinaReaderService` value. For Layer-based setup, prefer
+* {@link layer}.
+*/
+const make = (cfg) => Effect.map(HttpClient.HttpClient.asEffect(), (client) => ({ read: (request) => readImpl(cfg)(request).pipe(Effect.provideService(HttpClient.HttpClient, client)) }));
+/**
+* Layer registering both the provider-typed `JinaReader` tag and the generic
+* `WebRead` tag over one implementation. A `CommonReadRequest` is
+* structurally a `JinaReadRequest` with the extra knobs unset, so the generic
+* registration forwards directly.
+*/
+const layer = (cfg) => {
+	const typed = Layer.effect(JinaReader, make(cfg));
+	const generic = Layer.effect(WebRead, Effect.map(make(cfg), (s) => ({ read: (request) => s.read(request) })));
+	return Layer.merge(typed, generic);
+};
+//#endregion
+export { JinaReader, layer, make, JinaReader_exports as t };
+
+//# sourceMappingURL=JinaReader.mjs.map

package/dist/JinaReader.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"JinaReader.mjs","names":[],"sources":["../src/JinaReader.ts"],"sourcesContent":["import { Context, Duration, Effect, Layer, type Redacted, Schema } from \"effect\"\nimport { HttpClient, HttpClientRequest } from \"effect/unstable/http\"\nimport * as AiError from \"@effect-uai/core/AiError\"\nimport {\n  type CommonReadRequest,\n  type ReadResponse,\n  WebRead,\n  type WebReadService,\n} from \"@effect-uai/core/WebRead\"\nimport type { JinaEngine } from \"./models.js\"\n\n// ---------------------------------------------------------------------------\n// Public types\n// ---------------------------------------------------------------------------\n\n/**\n * Jina-typed read request. Widens {@link CommonReadRequest} with Jina\n * Reader's own header knobs; the common fields map onto Reader headers (see\n * the codec below).\n */\nexport type JinaReadRequest = CommonReadRequest & {\n  /** Fetching engine - see {@link JinaEngine}. */\n  readonly engine?: JinaEngine\n  /** CSS selector to extract instead of the whole page (`X-Target-Selector`). */\n  readonly targetSelector?: string\n  /** Bypass Jina's page cache (`X-No-Cache`). */\n  readonly noCache?: boolean\n  /** Append a deduplicated link list, populating {@link ReadResponse.links}. */\n  readonly withLinksSummary?: boolean\n}\n\nexport type JinaReaderService = {\n  readonly read: (request: JinaReadRequest) => Effect.Effect<ReadResponse, AiError.AiError>\n}\n\n/**\n * Provider-typed service tag. Yield this for Jina-typed reads (its own\n * Reader headers); yield the generic `WebRead` tag for provider-portable\n * code. Both are registered by {@link layer}.\n */\nexport class JinaReader extends Context.Service<JinaReader, JinaReaderService>()(\n  \"@betalyra/effect-uai/providers/jina/JinaReader\",\n) {}\n\nexport type Config = {\n  readonly apiKey: Redacted.Redacted\n  readonly baseUrl?: string\n}\n\n// ---------------------------------------------------------------------------\n// Codec - request headers\n// ---------------------------------------------------------------------------\n\n// Reader is header-driven: the URL is the path, everything else is a header.\n// `markdown` / `html` map one-to-one onto `X-Return-Format`.\nconst buildHeaders = (request: JinaReadRequest): Record<string, string> => {\n  const format = request.format ?? \"markdown\"\n  return {\n    Accept: \"application/json\",\n    \"X-Return-Format\": format,\n    ...(request.timeout !== undefined && {\n      // `X-Timeout` is whole seconds, capped at 180 by Jina.\n      \"X-Timeout\": String(Math.max(1, Math.round(Duration.toMillis(request.timeout) / 1000))),\n    }),\n    ...(request.engine !== undefined && { \"X-Engine\": request.engine }),\n    ...(request.targetSelector !== undefined && { \"X-Target-Selector\": request.targetSelector }),\n    ...(request.noCache === true && { \"X-No-Cache\": \"true\" }),\n    ...(request.withLinksSummary === true && { \"X-With-Links-Summary\": \"true\" }),\n  }\n}\n\n// ---------------------------------------------------------------------------\n// Codec - response\n// ---------------------------------------------------------------------------\n\nconst WireData = Schema.Struct({\n  title: Schema.optional(Schema.NullOr(Schema.String)),\n  url: Schema.optional(Schema.NullOr(Schema.String)),\n  // `content` carries markdown/text; `html` carries html. Exactly one is set,\n  // depending on the requested format.\n  content: Schema.optional(Schema.NullOr(Schema.String)),\n  html: Schema.optional(Schema.NullOr(Schema.String)),\n  // Present only with `X-With-Links-Summary`; a { anchorText: url } record.\n  links: Schema.optional(Schema.Unknown),\n})\n\nconst WireResponse = Schema.Struct({\n  code: Schema.optional(Schema.Number),\n  status: Schema.optional(Schema.Number),\n  data: WireData,\n})\ntype WireResponse = typeof WireResponse.Type\n\n// Jina reports links as a { text: url } record; take the URLs. Tolerate an\n// array shape too, and omit when empty.\nconst extractLinks = (links: unknown): ReadonlyArray<string> | undefined => {\n  const values = Array.isArray(links)\n    ? links\n    : links !== null && typeof links === \"object\"\n      ? Object.values(links as Record<string, unknown>)\n      : []\n  const urls = values.filter((v): v is string => typeof v === \"string\")\n  return urls.length > 0 ? urls : undefined\n}\n\nconst toResponse = (wire: WireResponse, request: JinaReadRequest): ReadResponse => {\n  const { data } = wire\n  const format = request.format ?? \"markdown\"\n  const content = (format === \"html\" ? data.html : data.content) ?? \"\"\n  const title = data.title ?? undefined\n  const links = extractLinks(data.links)\n  return {\n    url: data.url ?? request.url,\n    content,\n    ...(title !== undefined && { title }),\n    ...(links !== undefined && { links }),\n    raw: wire,\n  }\n}\n\n// ---------------------------------------------------------------------------\n// HTTP\n// ---------------------------------------------------------------------------\n\nconst transportFailure = (cause: unknown): AiError.AiError =>\n  new AiError.Unavailable({ provider: \"jina\", raw: cause })\n\nconst httpStatusError = (status: number, body: string): AiError.AiError => {\n  const provider = \"jina\"\n  const raw = body\n  if (status === 429) return new AiError.RateLimited({ provider, raw })\n  if (status === 408 || status === 504) return new AiError.Timeout({ provider, raw })\n  if (status === 401) return new AiError.AuthFailed({ provider, subtype: \"auth\", raw })\n  if (status === 403) return new AiError.AuthFailed({ provider, subtype: \"permission\", raw })\n  if (status === 402) return new AiError.AuthFailed({ provider, subtype: \"billing\", raw })\n  if (status >= 500) return new AiError.Unavailable({ provider, status, raw })\n  return new AiError.InvalidRequest({ provider, raw })\n}\n\nconst baseUrl = (cfg: Config): string => cfg.baseUrl ?? \"https://r.jina.ai\"\n\nconst getRead = (\n  cfg: Config,\n  request: JinaReadRequest,\n): Effect.Effect<WireResponse, AiError.AiError, HttpClient.HttpClient> =>\n  Effect.gen(function* () {\n    const client = yield* HttpClient.HttpClient\n    // The target URL is appended as the path: `https://r.jina.ai/<url>`.\n    const httpRequest = HttpClientRequest.get(`${baseUrl(cfg)}/${request.url}`).pipe(\n      HttpClientRequest.bearerToken(cfg.apiKey),\n      HttpClientRequest.setHeaders(buildHeaders(request)),\n    )\n    const response = yield* client.execute(httpRequest).pipe(Effect.mapError(transportFailure))\n    if (response.status >= 400) {\n      const text = yield* response.text.pipe(Effect.orElseSucceed(() => \"\"))\n      return yield* httpStatusError(response.status, text)\n    }\n    const json = yield* response.json.pipe(Effect.mapError(transportFailure))\n    return yield* Schema.decodeUnknownEffect(WireResponse)(json).pipe(\n      Effect.mapError(transportFailure),\n    )\n  })\n\nconst readImpl =\n  (cfg: Config) =>\n  (request: JinaReadRequest): Effect.Effect<ReadResponse, AiError.AiError, HttpClient.HttpClient> =>\n    Effect.map(getRead(cfg, request), (wire) => toResponse(wire, request))\n\n// ---------------------------------------------------------------------------\n// Constructors\n// ---------------------------------------------------------------------------\n\n/**\n * Build a `JinaReaderService` value. For Layer-based setup, prefer\n * {@link layer}.\n */\nexport const make = (cfg: Config): Effect.Effect<JinaReaderService, never, HttpClient.HttpClient> =>\n  Effect.map(HttpClient.HttpClient.asEffect(), (client) => ({\n    read: (request) =>\n      readImpl(cfg)(request).pipe(Effect.provideService(HttpClient.HttpClient, client)),\n  }))\n\n/**\n * Layer registering both the provider-typed `JinaReader` tag and the generic\n * `WebRead` tag over one implementation. A `CommonReadRequest` is\n * structurally a `JinaReadRequest` with the extra knobs unset, so the generic\n * registration forwards directly.\n */\nexport const layer = (\n  cfg: Config,\n): Layer.Layer<JinaReader | WebRead, never, HttpClient.HttpClient> => {\n  const typed = Layer.effect(JinaReader, make(cfg))\n  const generic = Layer.effect(\n    WebRead,\n    Effect.map(make(cfg), (s): WebReadService => ({ read: (request) => s.read(request) })),\n  )\n  return Layer.merge(typed, generic)\n}\n"],"mappings":";;;;;;;;;;;;;;;;AAwCA,IAAa,aAAb,cAAgC,QAAQ,SAAwC,CAC9E,iDACD,CAAC;AAaF,MAAM,gBAAgB,YAAqD;AAEzE,QAAO;EACL,QAAQ;EACR,mBAHa,QAAQ,UAAU;EAI/B,GAAI,QAAQ,YAAY,KAAA,KAAa,EAEnC,aAAa,OAAO,KAAK,IAAI,GAAG,KAAK,MAAM,SAAS,SAAS,QAAQ,QAAQ,GAAG,IAAK,CAAC,CAAC,EACxF;EACD,GAAI,QAAQ,WAAW,KAAA,KAAa,EAAE,YAAY,QAAQ,QAAQ;EAClE,GAAI,QAAQ,mBAAmB,KAAA,KAAa,EAAE,qBAAqB,QAAQ,gBAAgB;EAC3F,GAAI,QAAQ,YAAY,QAAQ,EAAE,cAAc,QAAQ;EACxD,GAAI,QAAQ,qBAAqB,QAAQ,EAAE,wBAAwB,QAAQ;EAC5E;;AAOH,MAAM,WAAW,OAAO,OAAO;CAC7B,OAAO,OAAO,SAAS,OAAO,OAAO,OAAO,OAAO,CAAC;CACpD,KAAK,OAAO,SAAS,OAAO,OAAO,OAAO,OAAO,CAAC;CAGlD,SAAS,OAAO,SAAS,OAAO,OAAO,OAAO,OAAO,CAAC;CACtD,MAAM,OAAO,SAAS,OAAO,OAAO,OAAO,OAAO,CAAC;CAEnD,OAAO,OAAO,SAAS,OAAO,QAAQ;CACvC,CAAC;AAEF,MAAM,eAAe,OAAO,OAAO;CACjC,MAAM,OAAO,SAAS,OAAO,OAAO;CACpC,QAAQ,OAAO,SAAS,OAAO,OAAO;CACtC,MAAM;CACP,CAAC;AAKF,MAAM,gBAAgB,UAAsD;CAM1E,MAAM,QALS,MAAM,QAAQ,MAAM,GAC/B,QACA,UAAU,QAAQ,OAAO,UAAU,WACjC,OAAO,OAAO,MAAiC,GAC/C,EAAE,EACY,QAAQ,MAAmB,OAAO,MAAM,SAAS;AACrE,QAAO,KAAK,SAAS,IAAI,OAAO,KAAA;;AAGlC,MAAM,cAAc,MAAoB,YAA2C;CACjF,MAAM,EAAE,SAAS;CAEjB,MAAM,YADS,QAAQ,UAAU,gBACL,SAAS,KAAK,OAAO,KAAK,YAAY;CAClE,MAAM,QAAQ,KAAK,SAAS,KAAA;CAC5B,MAAM,QAAQ,aAAa,KAAK,MAAM;AACtC,QAAO;EACL,KAAK,KAAK,OAAO,QAAQ;EACzB;EACA,GAAI,UAAU,KAAA,KAAa,EAAE,OAAO;EACpC,GAAI,UAAU,KAAA,KAAa,EAAE,OAAO;EACpC,KAAK;EACN;;AAOH,MAAM,oBAAoB,UACxB,IAAI,QAAQ,YAAY;CAAE,UAAU;CAAQ,KAAK;CAAO,CAAC;AAE3D,MAAM,mBAAmB,QAAgB,SAAkC;CACzE,MAAM,WAAW;CACjB,MAAM,MAAM;AACZ,KAAI,WAAW,IAAK,QAAO,IAAI,QAAQ,YAAY;EAAE;EAAU;EAAK,CAAC;AACrE,KAAI,WAAW,OAAO,WAAW,IAAK,QAAO,IAAI,QAAQ,QAAQ;EAAE;EAAU;EAAK,CAAC;AACnF,KAAI,WAAW,IAAK,QAAO,IAAI,QAAQ,WAAW;EAAE;EAAU,SAAS;EAAQ;EAAK,CAAC;AACrF,KAAI,WAAW,IAAK,QAAO,IAAI,QAAQ,WAAW;EAAE;EAAU,SAAS;EAAc;EAAK,CAAC;AAC3F,KAAI,WAAW,IAAK,QAAO,IAAI,QAAQ,WAAW;EAAE;EAAU,SAAS;EAAW;EAAK,CAAC;AACxF,KAAI,UAAU,IAAK,QAAO,IAAI,QAAQ,YAAY;EAAE;EAAU;EAAQ;EAAK,CAAC;AAC5E,QAAO,IAAI,QAAQ,eAAe;EAAE;EAAU;EAAK,CAAC;;AAGtD,MAAM,WAAW,QAAwB,IAAI,WAAW;AAExD,MAAM,WACJ,KACA,YAEA,OAAO,IAAI,aAAa;CACtB,MAAM,SAAS,OAAO,WAAW;CAEjC,MAAM,cAAc,kBAAkB,IAAI,GAAG,QAAQ,IAAI,CAAC,GAAG,QAAQ,MAAM,CAAC,KAC1E,kBAAkB,YAAY,IAAI,OAAO,EACzC,kBAAkB,WAAW,aAAa,QAAQ,CAAC,CACpD;CACD,MAAM,WAAW,OAAO,OAAO,QAAQ,YAAY,CAAC,KAAK,OAAO,SAAS,iBAAiB,CAAC;AAC3F,KAAI,SAAS,UAAU,KAAK;EAC1B,MAAM,OAAO,OAAO,SAAS,KAAK,KAAK,OAAO,oBAAoB,GAAG,CAAC;AACtE,SAAO,OAAO,gBAAgB,SAAS,QAAQ,KAAK;;CAEtD,MAAM,OAAO,OAAO,SAAS,KAAK,KAAK,OAAO,SAAS,iBAAiB,CAAC;AACzE,QAAO,OAAO,OAAO,oBAAoB,aAAa,CAAC,KAAK,CAAC,KAC3D,OAAO,SAAS,iBAAiB,CAClC;EACD;AAEJ,MAAM,YACH,SACA,YACC,OAAO,IAAI,QAAQ,KAAK,QAAQ,GAAG,SAAS,WAAW,MAAM,QAAQ,CAAC;;;;;AAU1E,MAAa,QAAQ,QACnB,OAAO,IAAI,WAAW,WAAW,UAAU,GAAG,YAAY,EACxD,OAAO,YACL,SAAS,IAAI,CAAC,QAAQ,CAAC,KAAK,OAAO,eAAe,WAAW,YAAY,OAAO,CAAC,EACpF,EAAE;;;;;;;AAQL,MAAa,SACX,QACoE;CACpE,MAAM,QAAQ,MAAM,OAAO,YAAY,KAAK,IAAI,CAAC;CACjD,MAAM,UAAU,MAAM,OACpB,SACA,OAAO,IAAI,KAAK,IAAI,GAAG,OAAuB,EAAE,OAAO,YAAY,EAAE,KAAK,QAAQ,EAAE,EAAE,CACvF;AACD,QAAO,MAAM,MAAM,OAAO,QAAQ"}

package/dist/index.d.mts

@@ -1,3 +1,4 @@
-import { JinaEmbeddingModel } from "./models.mjs";
+import { JinaEmbeddingModel, JinaEngine } from "./models.mjs";
 import { t as JinaEmbedding_d_exports } from "./JinaEmbedding.mjs";
-export { JinaEmbedding_d_exports as JinaEmbedding, JinaEmbeddingModel };
+import { t as JinaReader_d_exports } from "./JinaReader.mjs";
+export { JinaEmbedding_d_exports as JinaEmbedding, JinaEmbeddingModel, JinaEngine, JinaReader_d_exports as JinaReader };

package/dist/index.mjs

@@ -1,3 +1,4 @@
 import { t as JinaEmbedding_exports } from "./JinaEmbedding.mjs";
+import { t as JinaReader_exports } from "./JinaReader.mjs";
 import "./models.mjs";
-export { JinaEmbedding_exports as JinaEmbedding };
+export { JinaEmbedding_exports as JinaEmbedding, JinaReader_exports as JinaReader };

package/dist/models.d.mts

@@ -19,6 +19,12 @@
  * Reference: https://jina.ai/embeddings/
  */
 type JinaEmbeddingModel = "jina-embeddings-v4" | "jina-embeddings-v5-text-small" | "jina-embeddings-v5-text-nano" | "jina-embeddings-v3" | "jina-clip-v2" | (string & {});
+/**
+ * Reader fetching engine (`X-Engine`). `browser` renders JavaScript,
+ * `curl` is a fast static fetch, `auto` picks per page. Provider-specific,
+ * so it lives on `JinaReadRequest`.
+ */
+type JinaEngine = "browser" | "curl" | "auto" | (string & {});
 //#endregion
-export { JinaEmbeddingModel };
+export { JinaEmbeddingModel, JinaEngine };
 //# sourceMappingURL=models.d.mts.map

package/dist/models.d.mts.map

@@ -1 +1 @@
-{"version":3,"file":"models.d.mts","names":[],"sources":["../src/models.ts"],"mappings":";;AAmBA;;;;;;;;;;;;;;;;;;KAAY,kBAAA"}
+{"version":3,"file":"models.d.mts","names":[],"sources":["../src/models.ts"],"mappings":";;AAmBA;;;;;AAcA;;;;;;;;;;;;;KAdY,kBAAA;;;;;;KAcA,UAAA"}

package/package.json

@@ -1,13 +1,14 @@
 {
   "name": "@effect-uai/jina",
-  "version": "0.9.0",
-  "description": "Jina AI provider for @effect-uai/core (embeddings, rerank).",
+  "version": "0.10.0",
+  "description": "Jina AI provider for @effect-uai/core (embeddings, rerank, web read).",
   "keywords": [
     "ai",
     "effect",
     "embeddings",
     "jina",
-    "rerank"
+    "rerank",
+    "web-read"
   ],
   "homepage": "https://github.com/betalyra/effect-uai/tree/main/packages/providers/jina",
   "bugs": {
@@ -37,6 +38,10 @@
     "./JinaEmbedding": {
       "types": "./dist/JinaEmbedding.d.mts",
       "import": "./dist/JinaEmbedding.mjs"
+    },
+    "./JinaReader": {
+      "types": "./dist/JinaReader.d.mts",
+      "import": "./dist/JinaReader.mjs"
     }
   },
   "publishConfig": {
@@ -45,7 +50,7 @@
   "devDependencies": {
     "effect": "4.0.0-beta.57",
     "typescript": "^6.0.3",
-    "@effect-uai/core": "0.9.0"
+    "@effect-uai/core": "0.10.0"
   },
   "peerDependencies": {
     "@effect-uai/core": ">=0.2.0 <1",

package/src/JinaReader.ts

@@ -0,0 +1,198 @@
+import { Context, Duration, Effect, Layer, type Redacted, Schema } from "effect"
+import { HttpClient, HttpClientRequest } from "effect/unstable/http"
+import * as AiError from "@effect-uai/core/AiError"
+import {
+  type CommonReadRequest,
+  type ReadResponse,
+  WebRead,
+  type WebReadService,
+} from "@effect-uai/core/WebRead"
+import type { JinaEngine } from "./models.js"
+
+// ---------------------------------------------------------------------------
+// Public types
+// ---------------------------------------------------------------------------
+
+/**
+ * Jina-typed read request. Widens {@link CommonReadRequest} with Jina
+ * Reader's own header knobs; the common fields map onto Reader headers (see
+ * the codec below).
+ */
+export type JinaReadRequest = CommonReadRequest & {
+  /** Fetching engine - see {@link JinaEngine}. */
+  readonly engine?: JinaEngine
+  /** CSS selector to extract instead of the whole page (`X-Target-Selector`). */
+  readonly targetSelector?: string
+  /** Bypass Jina's page cache (`X-No-Cache`). */
+  readonly noCache?: boolean
+  /** Append a deduplicated link list, populating {@link ReadResponse.links}. */
+  readonly withLinksSummary?: boolean
+}
+
+export type JinaReaderService = {
+  readonly read: (request: JinaReadRequest) => Effect.Effect<ReadResponse, AiError.AiError>
+}
+
+/**
+ * Provider-typed service tag. Yield this for Jina-typed reads (its own
+ * Reader headers); yield the generic `WebRead` tag for provider-portable
+ * code. Both are registered by {@link layer}.
+ */
+export class JinaReader extends Context.Service<JinaReader, JinaReaderService>()(
+  "@betalyra/effect-uai/providers/jina/JinaReader",
+) {}
+
+export type Config = {
+  readonly apiKey: Redacted.Redacted
+  readonly baseUrl?: string
+}
+
+// ---------------------------------------------------------------------------
+// Codec - request headers
+// ---------------------------------------------------------------------------
+
+// Reader is header-driven: the URL is the path, everything else is a header.
+// `markdown` / `html` map one-to-one onto `X-Return-Format`.
+const buildHeaders = (request: JinaReadRequest): Record<string, string> => {
+  const format = request.format ?? "markdown"
+  return {
+    Accept: "application/json",
+    "X-Return-Format": format,
+    ...(request.timeout !== undefined && {
+      // `X-Timeout` is whole seconds, capped at 180 by Jina.
+      "X-Timeout": String(Math.max(1, Math.round(Duration.toMillis(request.timeout) / 1000))),
+    }),
+    ...(request.engine !== undefined && { "X-Engine": request.engine }),
+    ...(request.targetSelector !== undefined && { "X-Target-Selector": request.targetSelector }),
+    ...(request.noCache === true && { "X-No-Cache": "true" }),
+    ...(request.withLinksSummary === true && { "X-With-Links-Summary": "true" }),
+  }
+}
+
+// ---------------------------------------------------------------------------
+// Codec - response
+// ---------------------------------------------------------------------------
+
+const WireData = Schema.Struct({
+  title: Schema.optional(Schema.NullOr(Schema.String)),
+  url: Schema.optional(Schema.NullOr(Schema.String)),
+  // `content` carries markdown/text; `html` carries html. Exactly one is set,
+  // depending on the requested format.
+  content: Schema.optional(Schema.NullOr(Schema.String)),
+  html: Schema.optional(Schema.NullOr(Schema.String)),
+  // Present only with `X-With-Links-Summary`; a { anchorText: url } record.
+  links: Schema.optional(Schema.Unknown),
+})
+
+const WireResponse = Schema.Struct({
+  code: Schema.optional(Schema.Number),
+  status: Schema.optional(Schema.Number),
+  data: WireData,
+})
+type WireResponse = typeof WireResponse.Type
+
+// Jina reports links as a { text: url } record; take the URLs. Tolerate an
+// array shape too, and omit when empty.
+const extractLinks = (links: unknown): ReadonlyArray<string> | undefined => {
+  const values = Array.isArray(links)
+    ? links
+    : links !== null && typeof links === "object"
+      ? Object.values(links as Record<string, unknown>)
+      : []
+  const urls = values.filter((v): v is string => typeof v === "string")
+  return urls.length > 0 ? urls : undefined
+}
+
+const toResponse = (wire: WireResponse, request: JinaReadRequest): ReadResponse => {
+  const { data } = wire
+  const format = request.format ?? "markdown"
+  const content = (format === "html" ? data.html : data.content) ?? ""
+  const title = data.title ?? undefined
+  const links = extractLinks(data.links)
+  return {
+    url: data.url ?? request.url,
+    content,
+    ...(title !== undefined && { title }),
+    ...(links !== undefined && { links }),
+    raw: wire,
+  }
+}
+
+// ---------------------------------------------------------------------------
+// HTTP
+// ---------------------------------------------------------------------------
+
+const transportFailure = (cause: unknown): AiError.AiError =>
+  new AiError.Unavailable({ provider: "jina", raw: cause })
+
+const httpStatusError = (status: number, body: string): AiError.AiError => {
+  const provider = "jina"
+  const raw = body
+  if (status === 429) return new AiError.RateLimited({ provider, raw })
+  if (status === 408 || status === 504) return new AiError.Timeout({ provider, raw })
+  if (status === 401) return new AiError.AuthFailed({ provider, subtype: "auth", raw })
+  if (status === 403) return new AiError.AuthFailed({ provider, subtype: "permission", raw })
+  if (status === 402) return new AiError.AuthFailed({ provider, subtype: "billing", raw })
+  if (status >= 500) return new AiError.Unavailable({ provider, status, raw })
+  return new AiError.InvalidRequest({ provider, raw })
+}
+
+const baseUrl = (cfg: Config): string => cfg.baseUrl ?? "https://r.jina.ai"
+
+const getRead = (
+  cfg: Config,
+  request: JinaReadRequest,
+): Effect.Effect<WireResponse, AiError.AiError, HttpClient.HttpClient> =>
+  Effect.gen(function* () {
+    const client = yield* HttpClient.HttpClient
+    // The target URL is appended as the path: `https://r.jina.ai/<url>`.
+    const httpRequest = HttpClientRequest.get(`${baseUrl(cfg)}/${request.url}`).pipe(
+      HttpClientRequest.bearerToken(cfg.apiKey),
+      HttpClientRequest.setHeaders(buildHeaders(request)),
+    )
+    const response = yield* client.execute(httpRequest).pipe(Effect.mapError(transportFailure))
+    if (response.status >= 400) {
+      const text = yield* response.text.pipe(Effect.orElseSucceed(() => ""))
+      return yield* httpStatusError(response.status, text)
+    }
+    const json = yield* response.json.pipe(Effect.mapError(transportFailure))
+    return yield* Schema.decodeUnknownEffect(WireResponse)(json).pipe(
+      Effect.mapError(transportFailure),
+    )
+  })
+
+const readImpl =
+  (cfg: Config) =>
+  (request: JinaReadRequest): Effect.Effect<ReadResponse, AiError.AiError, HttpClient.HttpClient> =>
+    Effect.map(getRead(cfg, request), (wire) => toResponse(wire, request))
+
+// ---------------------------------------------------------------------------
+// Constructors
+// ---------------------------------------------------------------------------
+
+/**
+ * Build a `JinaReaderService` value. For Layer-based setup, prefer
+ * {@link layer}.
+ */
+export const make = (cfg: Config): Effect.Effect<JinaReaderService, never, HttpClient.HttpClient> =>
+  Effect.map(HttpClient.HttpClient.asEffect(), (client) => ({
+    read: (request) =>
+      readImpl(cfg)(request).pipe(Effect.provideService(HttpClient.HttpClient, client)),
+  }))
+
+/**
+ * Layer registering both the provider-typed `JinaReader` tag and the generic
+ * `WebRead` tag over one implementation. A `CommonReadRequest` is
+ * structurally a `JinaReadRequest` with the extra knobs unset, so the generic
+ * registration forwards directly.
+ */
+export const layer = (
+  cfg: Config,
+): Layer.Layer<JinaReader | WebRead, never, HttpClient.HttpClient> => {
+  const typed = Layer.effect(JinaReader, make(cfg))
+  const generic = Layer.effect(
+    WebRead,
+    Effect.map(make(cfg), (s): WebReadService => ({ read: (request) => s.read(request) })),
+  )
+  return Layer.merge(typed, generic)
+}

package/src/index.ts

@@ -1,2 +1,3 @@
 export * as JinaEmbedding from "./JinaEmbedding.js"
+export * as JinaReader from "./JinaReader.js"
 export * from "./models.js"

package/src/models.ts

@@ -25,3 +25,15 @@ export type JinaEmbeddingModel =
   | "jina-clip-v2"
   // eslint-disable-next-line @typescript-eslint/ban-types
   | (string & {})
+
+/**
+ * Reader fetching engine (`X-Engine`). `browser` renders JavaScript,
+ * `curl` is a fast static fetch, `auto` picks per page. Provider-specific,
+ * so it lives on `JinaReadRequest`.
+ */
+export type JinaEngine =
+  | "browser"
+  | "curl"
+  | "auto"
+  // eslint-disable-next-line @typescript-eslint/ban-types
+  | (string & {})