@lunora/ai 1.0.0-alpha.104 → 1.0.0-alpha.106

package/dist/index.d.mts

@@ -1,5 +1,5 @@
-import { L as LunoraAiOptions, a as LunoraAi } from "./packem_shared/types.d-BNOWTjHf.mjs";
-export { A as AI_DEFAULT_EMBEDDING_MODEL_ENV, b as AI_DEFAULT_MODEL_ENV, c as AI_GATEWAY_ACCOUNT_ID_ENV, d as AI_GATEWAY_ID_ENV, e as AI_GATEWAY_METADATA_MAX_KEYS, f as AI_GATEWAY_TAGS_ENV, g as AI_GATEWAY_TOKEN_ENV, h as AI_PROXY_TOKEN_ENV, i as AI_PROXY_URL_ENV, type j as AiBindingLike, type k as AiGatewayMetadata, type l as AiGatewayOptions, type m as AiMetrics, type n as AiSpan, type o as AiTelemetry, type p as AiTracer, type E as EmbeddingModelInput, type M as ModelInput, type R as ResolvedAiGateway, type W as WorkersAiProviderLike, q as buildAiGatewayMetadataFields, r as readAiGatewayEnvTags, s as resolveAiGateway } from "./packem_shared/types.d-BNOWTjHf.mjs";
+import { L as LunoraAiOptions, a as LunoraAi } from "./packem_shared/types.d-D-WK1t2B.mjs";
+export { A as AI_DEFAULT_EMBEDDING_MODEL_ENV, b as AI_DEFAULT_MODEL_ENV, c as AI_GATEWAY_ACCOUNT_ID_ENV, d as AI_GATEWAY_ID_ENV, e as AI_GATEWAY_METADATA_MAX_KEYS, f as AI_GATEWAY_TAGS_ENV, g as AI_GATEWAY_TOKEN_ENV, h as AI_PROXY_TOKEN_ENV, i as AI_PROXY_URL_ENV, type j as AiBindingLike, type k as AiGatewayMetadata, type l as AiGatewayOptions, type m as AiMetrics, type n as AiModelOptions, type o as AiRunOptions, type p as AiSpan, type q as AiTelemetry, type r as AiTracer, type s as AiWebSearchItem, type t as AiWebSearchOptions, type u as AiWebSearchProvider, type v as AiWebSearchResult, type E as EmbeddingModelInput, type M as ModelInput, type R as ResolvedAiGateway, type W as WorkersAiProviderLike, w as buildAiGatewayMetadataFields, x as readAiGatewayEnvTags, y as resolveAiGateway } from "./packem_shared/types.d-D-WK1t2B.mjs";
 export { type EmbeddingModel, type LanguageModel, embed, embedMany, generateObject, generateText, hasToolCall, jsonSchema, streamObject, streamText, tool } from 'ai';
 export { createWorkersAI } from 'workers-ai-provider';
 /**

package/dist/index.d.ts

@@ -1,5 +1,5 @@
-import { L as LunoraAiOptions, a as LunoraAi } from "./packem_shared/types.d-BNOWTjHf.js";
-export { A as AI_DEFAULT_EMBEDDING_MODEL_ENV, b as AI_DEFAULT_MODEL_ENV, c as AI_GATEWAY_ACCOUNT_ID_ENV, d as AI_GATEWAY_ID_ENV, e as AI_GATEWAY_METADATA_MAX_KEYS, f as AI_GATEWAY_TAGS_ENV, g as AI_GATEWAY_TOKEN_ENV, h as AI_PROXY_TOKEN_ENV, i as AI_PROXY_URL_ENV, type j as AiBindingLike, type k as AiGatewayMetadata, type l as AiGatewayOptions, type m as AiMetrics, type n as AiSpan, type o as AiTelemetry, type p as AiTracer, type E as EmbeddingModelInput, type M as ModelInput, type R as ResolvedAiGateway, type W as WorkersAiProviderLike, q as buildAiGatewayMetadataFields, r as readAiGatewayEnvTags, s as resolveAiGateway } from "./packem_shared/types.d-BNOWTjHf.js";
+import { L as LunoraAiOptions, a as LunoraAi } from "./packem_shared/types.d-D-WK1t2B.js";
+export { A as AI_DEFAULT_EMBEDDING_MODEL_ENV, b as AI_DEFAULT_MODEL_ENV, c as AI_GATEWAY_ACCOUNT_ID_ENV, d as AI_GATEWAY_ID_ENV, e as AI_GATEWAY_METADATA_MAX_KEYS, f as AI_GATEWAY_TAGS_ENV, g as AI_GATEWAY_TOKEN_ENV, h as AI_PROXY_TOKEN_ENV, i as AI_PROXY_URL_ENV, type j as AiBindingLike, type k as AiGatewayMetadata, type l as AiGatewayOptions, type m as AiMetrics, type n as AiModelOptions, type o as AiRunOptions, type p as AiSpan, type q as AiTelemetry, type r as AiTracer, type s as AiWebSearchItem, type t as AiWebSearchOptions, type u as AiWebSearchProvider, type v as AiWebSearchResult, type E as EmbeddingModelInput, type M as ModelInput, type R as ResolvedAiGateway, type W as WorkersAiProviderLike, w as buildAiGatewayMetadataFields, x as readAiGatewayEnvTags, y as resolveAiGateway } from "./packem_shared/types.d-D-WK1t2B.js";
 export { type EmbeddingModel, type LanguageModel, embed, embedMany, generateObject, generateText, hasToolCall, jsonSchema, streamObject, streamText, tool } from 'ai';
 export { createWorkersAI } from 'workers-ai-provider';
 /**

package/dist/index.mjs

@@ -1 +1 @@
-import{default as _}from"./packem_shared/createAi-CdiXSCUA.mjs";import{AI_DEFAULT_EMBEDDING_MODEL_ENV as t,AI_DEFAULT_MODEL_ENV as a,AI_GATEWAY_ACCOUNT_ID_ENV as o,AI_GATEWAY_ID_ENV as r,AI_GATEWAY_METADATA_MAX_KEYS as T,AI_GATEWAY_TAGS_ENV as I,AI_GATEWAY_TOKEN_ENV as N,AI_PROXY_TOKEN_ENV as l,AI_PROXY_URL_ENV as m,buildAiGatewayMetadataFields as s,readAiGatewayEnvTags as D,resolveAiGateway as G}from"./packem_shared/AI_DEFAULT_EMBEDDING_MODEL_ENV-IwcHZiKj.mjs";import{DEFAULT_MODEL_PRICES as O,estimateModelCost as d,lookupModelPrice as i}from"./packem_shared/DEFAULT_MODEL_PRICES-CIJLs1Ah.mjs";import{embed as Y,embedMany as x,generateObject as L,generateText as c,hasToolCall as f,jsonSchema as p,streamObject as W,streamText as b,tool as n}from"ai";import{createWorkersAI as U}from"workers-ai-provider";export{t as AI_DEFAULT_EMBEDDING_MODEL_ENV,a as AI_DEFAULT_MODEL_ENV,o as AI_GATEWAY_ACCOUNT_ID_ENV,r as AI_GATEWAY_ID_ENV,T as AI_GATEWAY_METADATA_MAX_KEYS,I as AI_GATEWAY_TAGS_ENV,N as AI_GATEWAY_TOKEN_ENV,l as AI_PROXY_TOKEN_ENV,m as AI_PROXY_URL_ENV,O as DEFAULT_MODEL_PRICES,s as buildAiGatewayMetadataFields,_ as createAi,U as createWorkersAI,Y as embed,x as embedMany,d as estimateModelCost,L as generateObject,c as generateText,f as hasToolCall,p as jsonSchema,i as lookupModelPrice,D as readAiGatewayEnvTags,G as resolveAiGateway,W as streamObject,b as streamText,n as tool};
+import{default as _}from"./packem_shared/createAi-Cebawe9R.mjs";import{AI_DEFAULT_EMBEDDING_MODEL_ENV as t,AI_DEFAULT_MODEL_ENV as a,AI_GATEWAY_ACCOUNT_ID_ENV as o,AI_GATEWAY_ID_ENV as r,AI_GATEWAY_METADATA_MAX_KEYS as T,AI_GATEWAY_TAGS_ENV as I,AI_GATEWAY_TOKEN_ENV as N,AI_PROXY_TOKEN_ENV as l,AI_PROXY_URL_ENV as m,buildAiGatewayMetadataFields as s,readAiGatewayEnvTags as D,resolveAiGateway as G}from"./packem_shared/AI_DEFAULT_EMBEDDING_MODEL_ENV-IwcHZiKj.mjs";import{DEFAULT_MODEL_PRICES as O,estimateModelCost as d,lookupModelPrice as i}from"./packem_shared/DEFAULT_MODEL_PRICES-CIJLs1Ah.mjs";import{embed as Y,embedMany as x,generateObject as L,generateText as c,hasToolCall as f,jsonSchema as p,streamObject as W,streamText as b,tool as n}from"ai";import{createWorkersAI as U}from"workers-ai-provider";export{t as AI_DEFAULT_EMBEDDING_MODEL_ENV,a as AI_DEFAULT_MODEL_ENV,o as AI_GATEWAY_ACCOUNT_ID_ENV,r as AI_GATEWAY_ID_ENV,T as AI_GATEWAY_METADATA_MAX_KEYS,I as AI_GATEWAY_TAGS_ENV,N as AI_GATEWAY_TOKEN_ENV,l as AI_PROXY_TOKEN_ENV,m as AI_PROXY_URL_ENV,O as DEFAULT_MODEL_PRICES,s as buildAiGatewayMetadataFields,_ as createAi,U as createWorkersAI,Y as embed,x as embedMany,d as estimateModelCost,L as generateObject,c as generateText,f as hasToolCall,p as jsonSchema,i as lookupModelPrice,D as readAiGatewayEnvTags,G as resolveAiGateway,W as streamObject,b as streamText,n as tool};

package/dist/packem_shared/createAi-Cebawe9R.mjs

@@ -0,0 +1 @@
+import{createOpenAI as W}from"@ai-sdk/openai";import{LunoraError as f}from"@lunora/errors";import{createWorkersAI as $}from"workers-ai-provider";import{anthropic as B}from"workers-ai-provider/anthropic";import{openai as V}from"workers-ai-provider/openai";import{buildAiGatewayMetadataFields as U,readEnv as p,AI_DEFAULT_MODEL_ENV as N,AI_DEFAULT_EMBEDDING_MODEL_ENV as L,AI_PROXY_URL_ENV as _,AI_PROXY_TOKEN_ENV as R,readAiGatewayEnvTags as F,AI_GATEWAY_ID_ENV as j,warnIgnoredBindingToken as H}from"./AI_DEFAULT_EMBEDDING_MODEL_ENV-IwcHZiKj.mjs";import{wrapLanguageModel as Y}from"ai";import{m as q,r as D}from"./usage-GPnTSqR8.mjs";import"./DEFAULT_MODEL_PRICES-CIJLs1Ah.mjs";const K={setAttribute:()=>{},setAttributes:()=>{}},P=async(e,t)=>t(P,K),X=(e,{metrics:t,trace:n=P})=>{const o={"gen_ai.operation.name":"chat","gen_ai.request.model":e};return{wrapGenerate:async({doGenerate:a})=>n("ai.generate",async(E,m)=>{const A=await a();return D(e,A,m,t),A},o),wrapStream:async({doStream:a})=>{const E=(h,d)=>{const c=h.stream.getReader();let u={};return{...h,stream:new ReadableStream({cancel:async w=>{d({outcome:u}),await c.cancel(w)},pull:async w=>{let v;try{v=await c.read()}catch(y){d({error:y,failed:!0,outcome:u}),w.error(y);return}if(v.done){d({outcome:u}),w.close();return}v.value.type==="finish"&&(u={providerMetadata:v.value.providerMetadata,usage:v.value.usage}),w.enqueue(v.value)}})}},m=Promise.withResolvers(),A=Promise.withResolvers();return n("ai.stream",async(h,d)=>{m.resolve(E(await a(),A.resolve));const c=await A.promise;if(D(e,c.outcome,d,t),c.failed===!0)throw c.error},o).catch(m.reject),m.promise}}},O=(e,t,n)=>{let o=e;return t!==void 0&&typeof e!="string"&&(o=Y({middleware:X(n??q(e)??"unknown",t),model:e})),o},Z=[V,B],S=e=>!e.startsWith("@")&&e.includes("/"),b=e=>{throw new f("INTERNAL",`@lunora/ai: ${e} needs the \`AI\` binding (env.AI). Add an \`ai\` binding to wrangler.jsonc, or set ${_} to an OpenAI-compatible proxy for "<provider>/<model>" slugs.`)},x=()=>b("this model id");x.textEmbeddingModel=()=>b("this embedding model id");const J=3040,Q=/^(?:[A-Z]\w*:\s*)?3040\s*:/iu,z=e=>e?.code===J||e instanceof Error&&Q.test(e.message),ee="default",te=new Map([[401,"UNAUTHORIZED"],[403,"FORBIDDEN"],[404,"NOT_FOUND"],[408,"SERVICE_UNAVAILABLE"],[429,"RATE_LIMITED"]]),re=e=>te.get(e)??(e>=400&&e<500?"BAD_REQUEST":"SERVICE_UNAVAILABLE"),ae=300,oe=new Set(["127.0.0.1","[::1]","localhost"]),ne=e=>{const t=p(e,_);if(t===void 0)return;const n=p(e,R),o=URL.canParse(t)?new URL(t):void 0;let a;if(o===void 0?a=`${_} is not a valid URL`:n!==void 0&&o.protocol!=="https:"&&!oe.has(o.hostname)&&(a=`${_} (${o.origin}) is not HTTPS, so ${R} would travel in cleartext — use an https:// URL`),a!==void 0){const E=()=>{throw new f("INTERNAL",`@lunora/ai: ${a}`)};return{chat:E,embedding:E}}return W({apiKey:n??"",baseURL:t,name:"lunora-proxy"})},ie=(e,t)=>{const n=e===void 0?void 0:F(e);return n===void 0?t:{...t,tags:{...n,...t?.tags}}},se=(e,t,n)=>{const o=U(n);if(e!==void 0)return o!==void 0&&e.metadata===void 0?{...e,metadata:o}:e;if(t===void 0)return;const a=p(t,j);if(a!==void 0)return H(t),o===void 0?{id:a}:{id:a,metadata:o}},ye=e=>{const{binding:t,defaultEmbeddingModel:n,defaultModel:o,env:a,gateway:E,metadata:m,provider:A,telemetry:h}=e,d=ne(a),c=ie(a,m),u=se(E,a,c),w=E?.metadata===void 0?U(c):void 0,v=o??p(a,N),y=n??p(a,L),g=A??(t?$({binding:t,gateway:u,providers:Z}):x),C=(r,i)=>S(r)?d!==void 0?d.chat(r):w===void 0?g(r):g(r,{metadata:w}):i?.rejectIfBusy===void 0?g(r):g(r,{rejectIfBusy:i.rejectIfBusy}),k=(r,i)=>{const s=r??v;if(s===void 0||s==="")throw new f("INTERNAL",`@lunora/ai: no model supplied and no default configured — pass a model id, or set ${N} in the Worker env (wrangler \`vars\` / \`.dev.vars\`)`);return typeof s=="string"?O(C(s,i),h,s):O(s,h)},G=r=>{if(d!==void 0&&S(r))return d.embedding(r);const i=g.textEmbeddingModel;if(typeof i!="function")throw new f("INTERNAL","@lunora/ai: the Workers AI provider does not expose `textEmbeddingModel`; pass an AI SDK EmbeddingModel (e.g. from @ai-sdk/openai) to embed()");return i.call(g,r)};return{embeddingModel:r=>{if(typeof r=="object")return r;const i=r??y;if(!i)throw new f("INTERNAL",`@lunora/ai: no embedding model supplied and no default configured — pass an embedding model id or an AI SDK EmbeddingModel, or set ${L} in the Worker env (wrangler \`vars\` / \`.dev.vars\`)`);return G(i)},model:k,run:async(r,i,s)=>{if(!t)return b("ai.run");const I=u!==void 0&&s?.gateway===void 0?{...s,gateway:u}:s;try{return await t.run(r,i,I)}catch(l){throw z(l)?new f("RATE_LIMITED",`@lunora/ai: Workers AI has no free capacity for ${r} (error 3040) — retry later`,{cause:l}):l}},websearch:async(r,i)=>{if(!t)return b("ai.websearch");if(typeof t.websearch!="function")throw new f("NOT_IMPLEMENTED","@lunora/ai: this Workers runtime's `AI` binding has no websearch() — update wrangler / @cloudflare/vite-plugin to a release that ships the Web Search API");const{gatewayId:s,...I}=i??{},l=await t.websearch({...I,gatewayId:s??u?.id??ee,query:r});if(!l.ok){const M=(await l.text().catch(()=>"")).slice(0,ae);throw new f(re(l.status),`@lunora/ai: web search failed with HTTP ${String(l.status)}${M===""?"":`: ${M}`}`)}const T=await l.json().catch(()=>{});if(!Array.isArray(T?.items))throw new f("INTERNAL","@lunora/ai: web search returned a body that is not JSON with an `items` array");return T},workersai:g}};export{ye as default};

package/dist/packem_shared/{types.d-BNOWTjHf.d.mts → types.d-D-WK1t2B.d.mts}

@@ -188,6 +188,15 @@ declare const resolveAiGateway: (env: Record<string, unknown>, metadata?: AiGate
  */
 interface AiBindingLike {
   run: (model: string, inputs: Record<string, unknown>, options?: Record<string, unknown>) => Promise<unknown>;
+  /**
+   * Web Search API (beta). Optional because runtimes and test doubles that
+   * predate it have no such method; `ctx.ai.websearch` reports that instead of
+   * a bare `TypeError`.
+   */
+  websearch?: (input: Omit<AiWebSearchOptions, "gatewayId"> & {
+    gatewayId: string;
+    query: string;
+  }) => Promise<Response>;
 }
 /**
  * A Workers AI provider instance — the value returned by `createWorkersAI(...)`.
@@ -216,6 +225,36 @@ interface AiGatewayOptions {
    */
   metadata?: Record<string, string>;
 }
+/**
+ * Options for the raw `ctx.ai.run(...)` passthrough — the Workers AI binding's
+ * third argument. Unlisted keys are forwarded to the binding unchanged.
+ * @experimental
+ */
+interface AiRunOptions {
+  [key: string]: unknown;
+  /** Route this call through a Cloudflare AI Gateway. Defaults to the gateway `createAi` resolved. */
+  gateway?: AiGatewayOptions;
+  /**
+   * Fail at once instead of waiting in the Workers AI capacity queue when no
+   * capacity is free. The rejection surfaces as a `LunoraError` with code
+   * `RATE_LIMITED` (Workers AI error `3040`, HTTP 429).
+   */
+  rejectIfBusy?: boolean;
+}
+/**
+ * Per-call settings for `ctx.ai.model(...)`. Applied to Workers AI model ids
+ * (`@cf/…`) only; a gateway slug or a bring-your-own model ignores them.
+ * @experimental
+ */
+interface AiModelOptions {
+  /**
+   * Fail at once instead of waiting in the Workers AI capacity queue. The
+   * provider reports the rejection as an AI SDK `APICallError` with
+   * `statusCode: 429`, which the AI SDK retries unless the call sets
+   * `maxRetries: 0`.
+   */
+  rejectIfBusy?: boolean;
+}
 /**
  * Structural slice of the span handle `ctx.trace` hands its body — enough to
  * attach a model call's usage once it is known. Declared here rather than
@@ -324,6 +363,59 @@ interface LunoraAiOptions {
    */
   telemetry?: AiTelemetry;
 }
+/**
+ * A search provider the Web Search API brokers. Each one runs under Cloudflare's
+ * Zero Data Retention terms. A search bills a provider key stored on the gateway
+ * when there is one (the `default` alias unless `byokAlias` names another),
+ * otherwise AI Gateway credits at list price.
+ * @experimental
+ */
+type AiWebSearchProvider = "ceramic" | "exa" | "linkup";
+/**
+ * Options for `ctx.ai.websearch(...)`.
+ * @experimental
+ */
+interface AiWebSearchOptions {
+  /** Alias of a provider key stored on the gateway to bill instead of the `default` alias. Without a stored key, gateway credits pay. */
+  byokAlias?: string;
+  /**
+   * The AI Gateway that brokers and bills the search. Defaults to the gateway
+   * `ctx.ai` routes inference through (`LUNORA_AI_GATEWAY_ID`), else the
+   * account's `default` gateway.
+   */
+  gatewayId?: string;
+  /** Maximum results, 1–10. Defaults to 10. */
+  limit?: number;
+  /** Defaults to `"ceramic"`. */
+  provider?: AiWebSearchProvider;
+}
+/**
+ * One result. Web search is discovery only: a result describes a page, and
+ * reading the page is a separate `fetch` of its `url`. The optional fields are
+ * present only when the provider returns them.
+ * @experimental
+ */
+interface AiWebSearchItem {
+  description?: string;
+  faviconUrl?: string;
+  imageUrl?: string;
+  /** Naive (no timezone) ISO-8601 datetime, e.g. `"2025-11-30T04:39:48"`. */
+  lastModifiedDate?: string;
+  title: string;
+  url: string;
+}
+/**
+ * What `ctx.ai.websearch(...)` resolves to.
+ * @experimental
+ */
+interface AiWebSearchResult {
+  items: AiWebSearchItem[];
+  metadata: {
+    latencyMs: number;
+    query: string;
+    requestId: string;
+  };
+}
 /**
  * A model to run against. The AI SDK's {@link LanguageModel} already admits a
  * bare `string`, so this alias covers both arms of the provider-agnostic seam:
@@ -360,15 +452,22 @@ interface LunoraAi {
    * provider key in the app) or, with `LUNORA_AI_PROXY_URL` set, to that
    * OpenAI-compatible proxy; an object → passthrough.
    */
-  model: (model?: ModelInput) => LanguageModel;
+  model: (model?: ModelInput, options?: AiModelOptions) => LanguageModel;
   /**
    * Raw Workers AI binding passthrough (void-style `ai.run`). Bypasses the AI
    * SDK entirely — useful for Workers-AI-only model families (image, ASR,
    * translation) not surfaced through the provider. Throws if no binding was
    * supplied.
    */
-  run: (model: string, inputs: Record<string, unknown>, options?: Record<string, unknown>) => Promise<unknown>;
+  run: (model: string, inputs: Record<string, unknown>, options?: AiRunOptions) => Promise<unknown>;
+  /**
+   * Search the web through the binding's Web Search API (beta) to ground a
+   * response in live information. Billed to a stored provider key or AI Gateway
+   * credits (see {@link AiWebSearchProvider}). Throws if no
+   * binding was supplied, or if the runtime's binding has no `websearch()`.
+   */
+  websearch: (query: string, options?: AiWebSearchOptions) => Promise<AiWebSearchResult>;
   /** The underlying Workers AI provider — `ai.workersai("@cf/...")` for a raw model. */
   workersai: WorkersAiProviderLike;
 }
-export { AI_DEFAULT_EMBEDDING_MODEL_ENV as A, EmbeddingModelInput as E, LunoraAiOptions as L, ModelInput as M, ResolvedAiGateway as R, WorkersAiProviderLike as W, LunoraAi as a, AI_DEFAULT_MODEL_ENV as b, AI_GATEWAY_ACCOUNT_ID_ENV as c, AI_GATEWAY_ID_ENV as d, AI_GATEWAY_METADATA_MAX_KEYS as e, AI_GATEWAY_TAGS_ENV as f, AI_GATEWAY_TOKEN_ENV as g, AI_PROXY_TOKEN_ENV as h, AI_PROXY_URL_ENV as i, AiBindingLike as j, AiGatewayMetadata as k, AiGatewayOptions as l, AiMetrics as m, AiSpan as n, AiTelemetry as o, AiTracer as p, buildAiGatewayMetadataFields as q, readAiGatewayEnvTags as r, resolveAiGateway as s };
+export { AI_DEFAULT_EMBEDDING_MODEL_ENV as A, EmbeddingModelInput as E, LunoraAiOptions as L, ModelInput as M, ResolvedAiGateway as R, WorkersAiProviderLike as W, LunoraAi as a, AI_DEFAULT_MODEL_ENV as b, AI_GATEWAY_ACCOUNT_ID_ENV as c, AI_GATEWAY_ID_ENV as d, AI_GATEWAY_METADATA_MAX_KEYS as e, AI_GATEWAY_TAGS_ENV as f, AI_GATEWAY_TOKEN_ENV as g, AI_PROXY_TOKEN_ENV as h, AI_PROXY_URL_ENV as i, AiBindingLike as j, AiGatewayMetadata as k, AiGatewayOptions as l, AiMetrics as m, AiModelOptions as n, AiRunOptions as o, AiSpan as p, AiTelemetry as q, AiTracer as r, AiWebSearchItem as s, AiWebSearchOptions as t, AiWebSearchProvider as u, AiWebSearchResult as v, buildAiGatewayMetadataFields as w, readAiGatewayEnvTags as x, resolveAiGateway as y };

package/dist/packem_shared/{types.d-BNOWTjHf.d.ts → types.d-D-WK1t2B.d.ts}

@@ -188,6 +188,15 @@ declare const resolveAiGateway: (env: Record<string, unknown>, metadata?: AiGate
  */
 interface AiBindingLike {
   run: (model: string, inputs: Record<string, unknown>, options?: Record<string, unknown>) => Promise<unknown>;
+  /**
+   * Web Search API (beta). Optional because runtimes and test doubles that
+   * predate it have no such method; `ctx.ai.websearch` reports that instead of
+   * a bare `TypeError`.
+   */
+  websearch?: (input: Omit<AiWebSearchOptions, "gatewayId"> & {
+    gatewayId: string;
+    query: string;
+  }) => Promise<Response>;
 }
 /**
  * A Workers AI provider instance — the value returned by `createWorkersAI(...)`.
@@ -216,6 +225,36 @@ interface AiGatewayOptions {
    */
   metadata?: Record<string, string>;
 }
+/**
+ * Options for the raw `ctx.ai.run(...)` passthrough — the Workers AI binding's
+ * third argument. Unlisted keys are forwarded to the binding unchanged.
+ * @experimental
+ */
+interface AiRunOptions {
+  [key: string]: unknown;
+  /** Route this call through a Cloudflare AI Gateway. Defaults to the gateway `createAi` resolved. */
+  gateway?: AiGatewayOptions;
+  /**
+   * Fail at once instead of waiting in the Workers AI capacity queue when no
+   * capacity is free. The rejection surfaces as a `LunoraError` with code
+   * `RATE_LIMITED` (Workers AI error `3040`, HTTP 429).
+   */
+  rejectIfBusy?: boolean;
+}
+/**
+ * Per-call settings for `ctx.ai.model(...)`. Applied to Workers AI model ids
+ * (`@cf/…`) only; a gateway slug or a bring-your-own model ignores them.
+ * @experimental
+ */
+interface AiModelOptions {
+  /**
+   * Fail at once instead of waiting in the Workers AI capacity queue. The
+   * provider reports the rejection as an AI SDK `APICallError` with
+   * `statusCode: 429`, which the AI SDK retries unless the call sets
+   * `maxRetries: 0`.
+   */
+  rejectIfBusy?: boolean;
+}
 /**
  * Structural slice of the span handle `ctx.trace` hands its body — enough to
  * attach a model call's usage once it is known. Declared here rather than
@@ -324,6 +363,59 @@ interface LunoraAiOptions {
    */
   telemetry?: AiTelemetry;
 }
+/**
+ * A search provider the Web Search API brokers. Each one runs under Cloudflare's
+ * Zero Data Retention terms. A search bills a provider key stored on the gateway
+ * when there is one (the `default` alias unless `byokAlias` names another),
+ * otherwise AI Gateway credits at list price.
+ * @experimental
+ */
+type AiWebSearchProvider = "ceramic" | "exa" | "linkup";
+/**
+ * Options for `ctx.ai.websearch(...)`.
+ * @experimental
+ */
+interface AiWebSearchOptions {
+  /** Alias of a provider key stored on the gateway to bill instead of the `default` alias. Without a stored key, gateway credits pay. */
+  byokAlias?: string;
+  /**
+   * The AI Gateway that brokers and bills the search. Defaults to the gateway
+   * `ctx.ai` routes inference through (`LUNORA_AI_GATEWAY_ID`), else the
+   * account's `default` gateway.
+   */
+  gatewayId?: string;
+  /** Maximum results, 1–10. Defaults to 10. */
+  limit?: number;
+  /** Defaults to `"ceramic"`. */
+  provider?: AiWebSearchProvider;
+}
+/**
+ * One result. Web search is discovery only: a result describes a page, and
+ * reading the page is a separate `fetch` of its `url`. The optional fields are
+ * present only when the provider returns them.
+ * @experimental
+ */
+interface AiWebSearchItem {
+  description?: string;
+  faviconUrl?: string;
+  imageUrl?: string;
+  /** Naive (no timezone) ISO-8601 datetime, e.g. `"2025-11-30T04:39:48"`. */
+  lastModifiedDate?: string;
+  title: string;
+  url: string;
+}
+/**
+ * What `ctx.ai.websearch(...)` resolves to.
+ * @experimental
+ */
+interface AiWebSearchResult {
+  items: AiWebSearchItem[];
+  metadata: {
+    latencyMs: number;
+    query: string;
+    requestId: string;
+  };
+}
 /**
  * A model to run against. The AI SDK's {@link LanguageModel} already admits a
  * bare `string`, so this alias covers both arms of the provider-agnostic seam:
@@ -360,15 +452,22 @@ interface LunoraAi {
    * provider key in the app) or, with `LUNORA_AI_PROXY_URL` set, to that
    * OpenAI-compatible proxy; an object → passthrough.
    */
-  model: (model?: ModelInput) => LanguageModel;
+  model: (model?: ModelInput, options?: AiModelOptions) => LanguageModel;
   /**
    * Raw Workers AI binding passthrough (void-style `ai.run`). Bypasses the AI
    * SDK entirely — useful for Workers-AI-only model families (image, ASR,
    * translation) not surfaced through the provider. Throws if no binding was
    * supplied.
    */
-  run: (model: string, inputs: Record<string, unknown>, options?: Record<string, unknown>) => Promise<unknown>;
+  run: (model: string, inputs: Record<string, unknown>, options?: AiRunOptions) => Promise<unknown>;
+  /**
+   * Search the web through the binding's Web Search API (beta) to ground a
+   * response in live information. Billed to a stored provider key or AI Gateway
+   * credits (see {@link AiWebSearchProvider}). Throws if no
+   * binding was supplied, or if the runtime's binding has no `websearch()`.
+   */
+  websearch: (query: string, options?: AiWebSearchOptions) => Promise<AiWebSearchResult>;
   /** The underlying Workers AI provider — `ai.workersai("@cf/...")` for a raw model. */
   workersai: WorkersAiProviderLike;
 }
-export { AI_DEFAULT_EMBEDDING_MODEL_ENV as A, EmbeddingModelInput as E, LunoraAiOptions as L, ModelInput as M, ResolvedAiGateway as R, WorkersAiProviderLike as W, LunoraAi as a, AI_DEFAULT_MODEL_ENV as b, AI_GATEWAY_ACCOUNT_ID_ENV as c, AI_GATEWAY_ID_ENV as d, AI_GATEWAY_METADATA_MAX_KEYS as e, AI_GATEWAY_TAGS_ENV as f, AI_GATEWAY_TOKEN_ENV as g, AI_PROXY_TOKEN_ENV as h, AI_PROXY_URL_ENV as i, AiBindingLike as j, AiGatewayMetadata as k, AiGatewayOptions as l, AiMetrics as m, AiSpan as n, AiTelemetry as o, AiTracer as p, buildAiGatewayMetadataFields as q, readAiGatewayEnvTags as r, resolveAiGateway as s };
+export { AI_DEFAULT_EMBEDDING_MODEL_ENV as A, EmbeddingModelInput as E, LunoraAiOptions as L, ModelInput as M, ResolvedAiGateway as R, WorkersAiProviderLike as W, LunoraAi as a, AI_DEFAULT_MODEL_ENV as b, AI_GATEWAY_ACCOUNT_ID_ENV as c, AI_GATEWAY_ID_ENV as d, AI_GATEWAY_METADATA_MAX_KEYS as e, AI_GATEWAY_TAGS_ENV as f, AI_GATEWAY_TOKEN_ENV as g, AI_PROXY_TOKEN_ENV as h, AI_PROXY_URL_ENV as i, AiBindingLike as j, AiGatewayMetadata as k, AiGatewayOptions as l, AiMetrics as m, AiModelOptions as n, AiRunOptions as o, AiSpan as p, AiTelemetry as q, AiTracer as r, AiWebSearchItem as s, AiWebSearchOptions as t, AiWebSearchProvider as u, AiWebSearchResult as v, buildAiGatewayMetadataFields as w, readAiGatewayEnvTags as x, resolveAiGateway as y };

package/dist/rag/index.d.mts

@@ -1,5 +1,5 @@
 import { Tool } from 'ai';
-import { E as EmbeddingModelInput, a as LunoraAi } from "../packem_shared/types.d-BNOWTjHf.mjs";
+import { E as EmbeddingModelInput, a as LunoraAi } from "../packem_shared/types.d-D-WK1t2B.mjs";
 /**
  * Built-in fixed-window chunker: split into `size`-char windows overlapping by
  * `overlap` chars. Deliberately simple and deterministic — the zero-config

package/dist/rag/index.d.ts

@@ -1,5 +1,5 @@
 import { Tool } from 'ai';
-import { E as EmbeddingModelInput, a as LunoraAi } from "../packem_shared/types.d-BNOWTjHf.js";
+import { E as EmbeddingModelInput, a as LunoraAi } from "../packem_shared/types.d-D-WK1t2B.js";
 /**
  * Built-in fixed-window chunker: split into `size`-char windows overlapping by
  * `overlap` chars. Deliberately simple and deterministic — the zero-config

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lunora/ai",
-  "version": "1.0.0-alpha.104",
+  "version": "1.0.0-alpha.106",
   "description": "Workers AI helper for Lunora: provider-agnostic AI SDK access from functions, Workers AI by default",
   "keywords": [
     "ai",
@@ -55,7 +55,7 @@
   "dependencies": {
     "@ai-sdk/anthropic": "4.0.49",
     "@ai-sdk/openai": "4.0.60",
-    "@lunora/errors": "1.0.0-alpha.47",
+    "@lunora/errors": "1.0.0-alpha.49",
     "ai": "7.0.93",
     "workers-ai-provider": "4.0.0"
   },

package/dist/packem_shared/createAi-CdiXSCUA.mjs

@@ -1 +0,0 @@
-import{createOpenAI as G}from"@ai-sdk/openai";import{LunoraError as w}from"@lunora/errors";import{createWorkersAI as $}from"workers-ai-provider";import{anthropic as S}from"workers-ai-provider/anthropic";import{openai as W}from"workers-ai-provider/openai";import{buildAiGatewayMetadataFields as T,readEnv as A,AI_DEFAULT_MODEL_ENV as M,AI_DEFAULT_EMBEDDING_MODEL_ENV as _,AI_PROXY_URL_ENV as b,AI_PROXY_TOKEN_ENV as R,readAiGatewayEnvTags as V,AI_GATEWAY_ID_ENV as F,warnIgnoredBindingToken as K}from"./AI_DEFAULT_EMBEDDING_MODEL_ENV-IwcHZiKj.mjs";import{wrapLanguageModel as q}from"ai";import{m as Y,r as I}from"./usage-GPnTSqR8.mjs";import"./DEFAULT_MODEL_PRICES-CIJLs1Ah.mjs";const j={setAttribute:()=>{},setAttributes:()=>{}},O=async(e,t)=>t(O,j),B=(e,{metrics:t,trace:a=O})=>{const o={"gen_ai.operation.name":"chat","gen_ai.request.model":e};return{wrapGenerate:async({doGenerate:r})=>a("ai.generate",async(c,v)=>{const f=await r();return I(e,f,v,t),f},o),wrapStream:async({doStream:r})=>{const c=(g,s)=>{const d=g.stream.getReader();let u={};return{...g,stream:new ReadableStream({cancel:async l=>{s({outcome:u}),await d.cancel(l)},pull:async l=>{let m;try{m=await d.read()}catch(E){s({error:E,failed:!0,outcome:u}),l.error(E);return}if(m.done){s({outcome:u}),l.close();return}m.value.type==="finish"&&(u={providerMetadata:m.value.providerMetadata,usage:m.value.usage}),l.enqueue(m.value)}})}},v=Promise.withResolvers(),f=Promise.withResolvers();return a("ai.stream",async(g,s)=>{v.resolve(c(await r(),f.resolve));const d=await f.promise;if(I(e,d.outcome,s,t),d.failed===!0)throw d.error},o).catch(v.reject),v.promise}}},L=(e,t,a)=>{let o=e;return t!==void 0&&typeof e!="string"&&(o=q({middleware:B(a??Y(e)??"unknown",t),model:e})),o},H=[W,S],N=e=>!e.startsWith("@")&&e.includes("/"),y=e=>{throw new w("INTERNAL",`@lunora/ai: ${e} needs the \`AI\` binding (env.AI). Add an \`ai\` binding to wrangler.jsonc, or set ${b} to an OpenAI-compatible proxy for "<provider>/<model>" slugs.`)},x=()=>y("this model id");x.textEmbeddingModel=()=>y("this embedding model id");const X=new Set(["127.0.0.1","[::1]","localhost"]),C=e=>{const t=A(e,b);if(t===void 0)return;const a=A(e,R),o=URL.canParse(t)?new URL(t):void 0;let r;if(o===void 0?r=`${b} is not a valid URL`:a!==void 0&&o.protocol!=="https:"&&!X.has(o.hostname)&&(r=`${b} (${o.origin}) is not HTTPS, so ${R} would travel in cleartext — use an https:// URL`),r!==void 0){const c=()=>{throw new w("INTERNAL",`@lunora/ai: ${r}`)};return{chat:c,embedding:c}}return G({apiKey:a??"",baseURL:t,name:"lunora-proxy"})},z=(e,t)=>{const a=e===void 0?void 0:V(e);return a===void 0?t:{...t,tags:{...a,...t?.tags}}},J=(e,t,a)=>{const o=T(a);if(e!==void 0)return o!==void 0&&e.metadata===void 0?{...e,metadata:o}:e;if(t===void 0)return;const r=A(t,F);if(r!==void 0)return K(t),o===void 0?{id:r}:{id:r,metadata:o}},ce=e=>{const{binding:t,defaultEmbeddingModel:a,defaultModel:o,env:r,gateway:c,metadata:v,provider:f,telemetry:g}=e,s=C(r),d=z(r,v),u=J(c,r,d),l=c?.metadata===void 0?T(d):void 0,m=o??A(r,M),E=a??A(r,_),p=f??(t?$({binding:t,gateway:u,providers:H}):x),D=n=>N(n)?s!==void 0?s.chat(n):l===void 0?p(n):p(n,{metadata:l}):p(n),P=n=>{const i=n??m;if(i===void 0||i==="")throw new w("INTERNAL",`@lunora/ai: no model supplied and no default configured — pass a model id, or set ${M} in the Worker env (wrangler \`vars\` / \`.dev.vars\`)`);return typeof i=="string"?L(D(i),g,i):L(i,g)},U=n=>{if(s!==void 0&&N(n))return s.embedding(n);const i=p.textEmbeddingModel;if(typeof i!="function")throw new w("INTERNAL","@lunora/ai: the Workers AI provider does not expose `textEmbeddingModel`; pass an AI SDK EmbeddingModel (e.g. from @ai-sdk/openai) to embed()");return i.call(p,n)};return{embeddingModel:n=>{if(typeof n=="object")return n;const i=n??E;if(!i)throw new w("INTERNAL",`@lunora/ai: no embedding model supplied and no default configured — pass an embedding model id or an AI SDK EmbeddingModel, or set ${_} in the Worker env (wrangler \`vars\` / \`.dev.vars\`)`);return U(i)},model:P,run:async(n,i,h)=>{if(!t)return y("ai.run");const k=u!==void 0&&h?.gateway===void 0?{...h,gateway:u}:h;return t.run(n,i,k)},workersai:p}};export{ce as default};