murmur-rag 1.0.0 → 1.2.0

package/dist/lib/embed.js

@@ -1,8 +1,8 @@
 import {
   __chunks,
   mountMurmur2
-} from "./chunks/index-4at214g0.js";
-__chunks(import.meta.url,["3qtg95t7","4at214g0","h8nz6jsc","gp7czjd1","vb9smwta","kaeec41p","c1c79eh8","6re28wc1","10g918kt","1zk822m5","ynjdey8j"],[["./embed.js",1],["./chunks/index-4at214g0.js",7],["./chunks/retriever-h8nz6jsc.js",8],["./indexer/index.js",9],["./llm/index.js"],["./chunks/chunk-kaeec41p.js",9],["./chunks/murmur-c1c79eh8.js",10,8],["./chunks/index-6re28wc1.js"],["./chunks/index-10g918kt.js",7],["./chunks/index-1zk822m5.js"],["./chunks/index-ynjdey8j.js",8,7]],0);
+} from "./chunks/index-y3v46yjp.js";
+__chunks(import.meta.url,["jc4f9310","y3v46yjp","t36hd4vt","wxshtnm0","dj4321kh","vb9smwta","nb4hdy2t","kaeec41p","b77xqap3","8h9c8ekq","p86sy3y4","1zk822m5"],[["./embed.js",1],["./chunks/index-y3v46yjp.js",8],["./chunks/murmur-t36hd4vt.js",9,10],["./chunks/retriever-wxshtnm0.js",10],["./indexer/index.js",11],["./llm/index.js"],["./local/index.js"],["./chunks/chunk-kaeec41p.js",11],["./chunks/index-b77xqap3.js"],["./chunks/index-8h9c8ekq.js",10,8],["./chunks/index-p86sy3y4.js",8],["./chunks/index-1zk822m5.js"]],0);
 
 // src/embed.ts
 function optionsFromDataset(data, base) {
@@ -45,6 +45,8 @@ function optionsFromDataset(data, base) {
     o.storageKey = data.storageKey;
   if (data.open === "true")
     o.open = true;
+  if (data.local === "false")
+    o.local = false;
   return o;
 }
 function start() {

package/dist/lib/index.js

@@ -1,11 +1,11 @@
 import {
   passagesOf2,
   bm25Retriever2
-} from "./chunks/index-10g918kt.js";
+} from "./chunks/index-p86sy3y4.js";
 import {
   searchQuery2,
   createMurmur2
-} from "./chunks/index-ynjdey8j.js";
+} from "./chunks/index-8h9c8ekq.js";
 import {
   MAX_CHUNK2,
   splitText2,
@@ -21,9 +21,10 @@ import {
   formatSources2,
   SOURCES_RULE2,
   groundingRules2,
+  cleanAnswer2,
   citedSources2,
   passageHref2
-} from "./chunks/index-6re28wc1.js";
+} from "./chunks/index-b77xqap3.js";
 export {
   MAX_CHUNK2 as MAX_CHUNK,
   SOURCES_RULE2 as SOURCES_RULE,
@@ -31,6 +32,7 @@ export {
   buildIndex2 as buildIndex,
   chunkSections2 as chunkSections,
   citedSources2 as citedSources,
+  cleanAnswer2 as cleanAnswer,
   createMurmur2 as createMurmur,
   estimateTokens2 as estimateTokens,
   formatSources2 as formatSources,

package/dist/lib/indexer/index.js

@@ -44,6 +44,8 @@ function extractSections(doc, { root = DEFAULT_ROOT, exclude = DEFAULT_EXCLUDE }
     if (tag === "BR")
       cur.text += `
 `;
+    if ((tag === "TD" || tag === "TH") && !/(^|\n)\s*$/.test(cur.text))
+      cur.text += " | ";
     const block = BLOCK.test(tag);
     if (block)
       cur.text += `

package/dist/lib/local/index.js

@@ -0,0 +1,163 @@
+// src/local/index.ts
+var DEFAULT_LOCAL_MODEL = {
+  id: "onnx-community/LFM3-1.7B-ONNX",
+  dtype: "q4f16",
+  name: "LFM3 1.7B",
+  sizeMB: 1090
+};
+async function hasWebGPU() {
+  const gpu = globalThis.navigator?.gpu;
+  if (!gpu)
+    return false;
+  try {
+    return !!await gpu.requestAdapter();
+  } catch {
+    return false;
+  }
+}
+function chatMessages(prompt) {
+  return prompt.flatMap((m) => {
+    if (m.role === "system")
+      return [{ role: "system", content: m.content }];
+    if (m.role === "tool")
+      return [];
+    const text = m.content.map((part) => part.type === "text" ? part.text : "").join("").trim();
+    return text ? [{ role: m.role, content: text }] : [];
+  });
+}
+var loaded = new Map;
+var listeners = new Map;
+function loadLocalModel(spec = DEFAULT_LOCAL_MODEL, onProgress) {
+  const key = `${spec.id}@${spec.dtype}`;
+  const set = listeners.get(key) ?? new Set;
+  listeners.set(key, set);
+  if (onProgress)
+    set.add(onProgress);
+  let pending = loaded.get(key);
+  if (!pending) {
+    pending = (async () => {
+      if (!await hasWebGPU())
+        throw new LocalModelError("webgpu", "This browser has no WebGPU.");
+      const { pipeline } = await import("@huggingface/transformers");
+      const generator = await pipeline("text-generation", spec.id, {
+        dtype: spec.dtype,
+        device: "webgpu",
+        progress_callback: (info) => {
+          if (info.status !== "progress_total")
+            return;
+          const p = {
+            loaded: info.loaded ?? 0,
+            total: info.total ?? 0,
+            progress: Math.min(1, (info.progress ?? 0) / 100)
+          };
+          for (const listener of set)
+            listener(p);
+        }
+      });
+      return generator;
+    })();
+    pending.catch(() => loaded.delete(key));
+    loaded.set(key, pending);
+  }
+  return pending;
+}
+
+class LocalModelError extends Error {
+  code;
+  constructor(code, message) {
+    super(message);
+    this.code = code;
+    this.name = "LocalModelError";
+  }
+}
+var usage = (input, output) => ({
+  inputTokens: { total: input, noCache: input, cacheRead: undefined, cacheWrite: undefined },
+  outputTokens: { total: output, text: output, reasoning: undefined }
+});
+function localModel(options = {}) {
+  const spec = options.spec ?? DEFAULT_LOCAL_MODEL;
+  async function run(call, onText) {
+    const generator = await loadLocalModel(spec, options.onProgress);
+    const { TextStreamer, InterruptableStoppingCriteria, StoppingCriteriaList } = await import("@huggingface/transformers");
+    const stop = new InterruptableStoppingCriteria;
+    const criteria = new StoppingCriteriaList;
+    criteria.push(stop);
+    const onAbort = () => stop.interrupt();
+    call.abortSignal?.addEventListener("abort", onAbort, { once: true });
+    let tokens = 0;
+    const streamer = new TextStreamer(generator.tokenizer, {
+      skip_prompt: true,
+      skip_special_tokens: true,
+      decode_kwargs: { clean_up_tokenization_spaces: false },
+      callback_function: (text) => {
+        if (text)
+          onText(text);
+      },
+      token_callback_function: () => {
+        tokens++;
+      }
+    });
+    try {
+      await generator(chatMessages(call.prompt), {
+        max_new_tokens: call.maxOutputTokens ?? options.maxNewTokens ?? 512,
+        do_sample: true,
+        temperature: call.temperature ?? 0.1,
+        top_k: 50,
+        repetition_penalty: 1.1,
+        streamer,
+        stopping_criteria: criteria,
+        tokenizer_encode_kwargs: { enable_thinking: false }
+      });
+    } finally {
+      call.abortSignal?.removeEventListener("abort", onAbort);
+    }
+    return { tokens, interrupted: stop.interrupted };
+  }
+  return {
+    specificationVersion: "v4",
+    provider: "murmur-local",
+    modelId: spec.id,
+    supportedUrls: {},
+    async doGenerate(call) {
+      let text = "";
+      const { tokens, interrupted } = await run(call, (t) => {
+        text += t;
+      });
+      return {
+        content: [{ type: "text", text }],
+        finishReason: { unified: interrupted ? "other" : "stop", raw: undefined },
+        usage: usage(0, tokens),
+        warnings: []
+      };
+    },
+    async doStream(call) {
+      const stream = new ReadableStream({
+        start(controller) {
+          controller.enqueue({ type: "stream-start", warnings: [] });
+          controller.enqueue({ type: "text-start", id: "0" });
+          run(call, (delta) => controller.enqueue({ type: "text-delta", id: "0", delta })).then(({ tokens, interrupted }) => {
+            controller.enqueue({ type: "text-end", id: "0" });
+            controller.enqueue({
+              type: "finish",
+              finishReason: { unified: interrupted ? "other" : "stop", raw: undefined },
+              usage: usage(0, tokens)
+            });
+            controller.close();
+          }, (error) => {
+            controller.enqueue({ type: "error", error });
+            controller.close();
+          });
+        }
+      });
+      return { stream };
+    }
+  };
+}
+export {
+  DEFAULT_LOCAL_MODEL,
+  LocalModelError,
+  chatMessages,
+  hasWebGPU,
+  loadLocalModel,
+  localModel
+};

package/dist/lib/rag/index.js

@@ -1,25 +1,26 @@
 import"../chunks/index-fp1p1byg.js";
-import {
-  MAX_CHUNK2,
-  splitText2,
-  slugify2,
-  chunkSections2,
-  markdownSections2,
-  buildIndex2,
-  isMurmurIndex2
-} from "../chunks/index-1zk822m5.js";
 import {
   passagesOf2,
   bm25Retriever2
-} from "../chunks/index-10g918kt.js";
+} from "../chunks/index-p86sy3y4.js";
 import {
   estimateTokens2,
   formatSources2,
   SOURCES_RULE2,
   groundingRules2,
+  cleanAnswer2,
   citedSources2,
   passageHref2
-} from "../chunks/index-6re28wc1.js";
+} from "../chunks/index-b77xqap3.js";
+import {
+  MAX_CHUNK2,
+  splitText2,
+  slugify2,
+  chunkSections2,
+  markdownSections2,
+  buildIndex2,
+  isMurmurIndex2
+} from "../chunks/index-1zk822m5.js";
 export {
   MAX_CHUNK2 as MAX_CHUNK,
   SOURCES_RULE2 as SOURCES_RULE,
@@ -27,6 +28,7 @@ export {
   buildIndex2 as buildIndex,
   chunkSections2 as chunkSections,
   citedSources2 as citedSources,
+  cleanAnswer2 as cleanAnswer,
   estimateTokens2 as estimateTokens,
   formatSources2 as formatSources,
   groundingRules2 as groundingRules,

package/dist/lib/widget/index.js

@@ -8,8 +8,8 @@ import {
   MurmurChat2,
   defineMurmurChat2,
   mountMurmur2
-} from "../chunks/index-4at214g0.js";
-__chunks(import.meta.url,["1ykj7t4p","4at214g0","h8nz6jsc","gp7czjd1","vb9smwta","kaeec41p","c1c79eh8","6re28wc1","10g918kt","1zk822m5","ynjdey8j"],[["./index.js",1],["../chunks/index-4at214g0.js",7],["../chunks/retriever-h8nz6jsc.js",8],["../indexer/index.js",9],["../llm/index.js"],["../chunks/chunk-kaeec41p.js",9],["../chunks/murmur-c1c79eh8.js",10,8],["../chunks/index-6re28wc1.js"],["../chunks/index-10g918kt.js",7],["../chunks/index-1zk822m5.js"],["../chunks/index-ynjdey8j.js",8,7]],0);
+} from "../chunks/index-y3v46yjp.js";
+__chunks(import.meta.url,["jzqx319b","y3v46yjp","t36hd4vt","wxshtnm0","dj4321kh","vb9smwta","nb4hdy2t","kaeec41p","b77xqap3","8h9c8ekq","p86sy3y4","1zk822m5"],[["./index.js",1],["../chunks/index-y3v46yjp.js",8],["../chunks/murmur-t36hd4vt.js",9,10],["../chunks/retriever-wxshtnm0.js",10],["../indexer/index.js",11],["../llm/index.js"],["../local/index.js"],["../chunks/chunk-kaeec41p.js",11],["../chunks/index-b77xqap3.js"],["../chunks/index-8h9c8ekq.js",10,8],["../chunks/index-p86sy3y4.js",8],["../chunks/index-1zk822m5.js"]],0);
 export {
   MESSAGES2 as MESSAGES,
   MurmurChat2 as MurmurChat,

package/dist/types/local/index.d.ts

@@ -0,0 +1,56 @@
+import type { LanguageModelV4, LanguageModelV4Prompt } from "@ai-sdk/provider";
+/** A model that runs in the page on WebGPU, through Transformers.js. */
+export interface LocalModelSpec {
+    /** Hugging Face repository of the ONNX weights. */
+    id: string;
+    /** Weights variant: q4f16 is the one meant for WebGPU. */
+    dtype: string;
+    /** Display name. */
+    name: string;
+    /** Approximate download, in megabytes, shown before the visitor agrees to it. */
+    sizeMB: number;
+}
+/**
+ * LFM3-1.7B by Liquid AI (October 2026), 4-bit weights with fp16 activations: the smallest
+ * text model released after mid-2026 with weights made for WebGPU. About 1.1 GB, downloaded
+ * once and kept in the browser's cache.
+ */
+export declare const DEFAULT_LOCAL_MODEL: LocalModelSpec;
+export interface LoadProgress {
+    /** Bytes loaded so far, across every file of the model. */
+    loaded: number;
+    /** Total bytes, once known. */
+    total: number;
+    /** 0 to 1. */
+    progress: number;
+}
+export interface LocalModelOptions {
+    /** Which model to run. Default {@link DEFAULT_LOCAL_MODEL}. */
+    spec?: LocalModelSpec;
+    /** Called while the weights download. */
+    onProgress?: (progress: LoadProgress) => void;
+    /** Longest answer, in tokens. Default 512. */
+    maxNewTokens?: number;
+}
+/** Whether this browser exposes a WebGPU adapter. */
+export declare function hasWebGPU(): Promise<boolean>;
+/** The chat messages a Transformers.js template takes, from an AI SDK prompt (text only). */
+export declare function chatMessages(prompt: LanguageModelV4Prompt): {
+    role: string;
+    content: string;
+}[];
+type Generator = ((messages: unknown, options: Record<string, unknown>) => Promise<unknown>) & {
+    tokenizer: unknown;
+};
+/** Downloads (or reads from cache) and compiles the model; later calls share the same load. */
+export declare function loadLocalModel(spec?: LocalModelSpec, onProgress?: (progress: LoadProgress) => void): Promise<Generator>;
+export declare class LocalModelError extends Error {
+    readonly code: "webgpu" | "load";
+    constructor(code: "webgpu" | "load", message: string);
+}
+/**
+ * An AI SDK language model that runs in the page: pass it to `createMurmur({ model })`, or let
+ * the widget create it with `local: true`. Nothing leaves the browser once the weights are cached.
+ */
+export declare function localModel(options?: LocalModelOptions): LanguageModelV4;
+export {};

package/dist/types/rag/ground.d.ts

@@ -6,6 +6,13 @@ export declare function formatSources(passages: readonly Passage[]): string;
 export declare const SOURCES_RULE: string;
 /** The rules every answer follows; the site owner's prompt comes before them. */
 export declare function groundingRules(language?: string): string;
+/**
+ * An answer as a visitor should read it, whatever the model: reasoning blocks and copied
+ * source delimiters removed (also half-written ones while streaming), citation variants
+ * (`[1, 2]`, `[Source 1]`, `【1】`) written as `[1][2]`, and a trailing "Sources: 1, 2"
+ * line dropped, since the interface lists the sources itself.
+ */
+export declare function cleanAnswer(text: string): string;
 /** The source numbers an answer cites, in order of first use, keeping only those that were given. */
 export declare function citedSources(answer: string, count: number): number[];
 /** Link to a passage: its page, at its heading. */

package/dist/types/rag/index.d.ts

@@ -1,4 +1,4 @@
 export { buildIndex, chunkSections, isMurmurIndex, MAX_CHUNK, markdownSections, type Section, slugify, splitText, } from "./chunk.js";
-export { citedSources, estimateTokens, formatSources, groundingRules, passageHref, SOURCES_RULE, } from "./ground.js";
+export { citedSources, cleanAnswer, estimateTokens, formatSources, groundingRules, passageHref, SOURCES_RULE, } from "./ground.js";
 export { bm25Retriever, passagesOf, type Retriever, type SearchOptions } from "./retriever.js";
 export type { Chunk, IndexDoc, MurmurIndex, Passage } from "./types.js";

package/dist/types/widget/element.d.ts

@@ -16,6 +16,7 @@ export declare class MurmurChat extends Base {
     private options;
     private session;
     private t;
+    private uiLang;
     private readonly root;
     private launcher;
     private panel;
@@ -48,6 +49,8 @@ export declare class MurmurChat extends Base {
     /** Draws an answer: its text with citation rings, then the sources it cites. */
     private fillAnswer;
     private submit;
+    /** Asks the visitor before downloading the in-browser model, or says why it cannot run. */
+    private showConsent;
     /** The visitor's model settings, drawn from llm-connect's store. */
     private showSetup;
     private hideSetup;

package/dist/types/widget/i18n.d.ts

@@ -40,6 +40,20 @@ export interface Messages {
     searchModels: string;
     recommended: string;
     otherProviders: string;
+    localOption: string;
+    localOptionHint: string;
+    localActive: string;
+    orKey: string;
+    useKey: string;
+    localConsent: string;
+    localDownload: string;
+    localLoading: string;
+    localLoadingLabel: string;
+    localPreparing: string;
+    localNoWebGPU: string;
+    localLoadFailed: string;
+    gigabytes: string;
+    megabytes: string;
 }
 export declare const MESSAGES: Record<string, Messages>;
 /** The first supported language among the candidates (`fr-CA` matches `fr`), else English. */

package/dist/types/widget/session.d.ts

@@ -1,6 +1,7 @@
 import type { LanguageModel } from "ai";
 import type { Turn } from "../core/murmur.js";
 import type { LlmConfig, LlmSettings, LlmSettingsStore } from "../llm/index.js";
+import type { LoadProgress, LocalModelSpec } from "../local/index.js";
 import type { Retriever } from "../rag/retriever.js";
 import type { MurmurIndex, Passage } from "../rag/types.js";
 import type { Messages } from "./i18n.js";
@@ -29,13 +30,20 @@ export interface WidgetOptions {
     storageKey?: string;
     /** Open the panel on load. */
     open?: boolean;
+    /**
+     * The model that runs in the visitor's browser on WebGPU, the default way to answer when there
+     * is no proxy: LFM3 1.7B, about 1.1 GB downloaded once after the visitor agrees. Visitors can
+     * still connect their own key in the AI settings. Pass your own spec to run another model, or
+     * `false` to offer only the visitor's key.
+     */
+    local?: boolean | LocalModelSpec;
 }
 export interface ChatTurn extends Turn {
     passages?: Pick<Passage, "url" | "anchor" | "title" | "heading">[];
     cited?: number[];
     error?: string;
 }
-type Mode = "custom" | "proxy" | "visitor";
+type Mode = "custom" | "local" | "proxy" | "visitor";
 /** Everything the widget does except drawing: index, model settings, history. */
 export declare class ChatSession {
     readonly options: WidgetOptions;
@@ -43,8 +51,29 @@ export declare class ChatSession {
     private engine;
     private storePromise;
     private readonly historyKey;
+    private readonly key;
+    /** Called while the in-browser model downloads. */
+    onProgress: ((progress: LoadProgress) => void) | null;
     constructor(options: WidgetOptions);
+    /**
+     * How this session answers: your own model, then the owner's proxy, then the visitor's own
+     * key if they chose one, else the model in the browser.
+     */
     get mode(): Mode;
+    /** Whether the in-browser model is offered: not with your own model or a proxy, nor `local: false`. */
+    get localAllowed(): boolean;
+    /**
+     * Before a question: a visitor who saved a key before the in-browser model became the default,
+     * and never chose between the two, keeps their key.
+     */
+    settleChoice(): Promise<void>;
+    /** The in-browser model this session would run. */
+    localSpec(): Promise<LocalModelSpec>;
+    /** The visitor picks the in-browser model, or their own key. */
+    useLocal(on: boolean): void;
+    /** Whether the visitor must still agree to download the in-browser model. */
+    get needsDownloadConsent(): boolean;
+    agreeToDownload(): void;
     persist(): void;
     clear(): void;
     /** The visitor's settings form, created on first use. */
@@ -52,6 +81,8 @@ export declare class ChatSession {
     /** The settings to call, or null when the visitor has to set up a model first. */
     settings(): Promise<LlmSettings | null>;
     needsSetup(): Promise<boolean>;
+    /** Whether a model can run in this browser at all. */
+    canRunLocally(): Promise<boolean>;
     private loadIndex;
     private createEngine;
     /** Asks a question; `onText` receives the answer so far. Resolves to the finished turn. */

package/dist/types/widget/styles.d.ts

@@ -3,4 +3,4 @@
  * custom properties on the element: --murmur-accent, --murmur-accent-ink, --murmur-font,
  * --murmur-radius, --murmur-z.
  */
-export declare const STYLES = "\n:host {\n  --accent: var(--murmur-accent, #211c17);\n  --accent-ink: var(--murmur-accent-ink, #f6efe0);\n  --lamp: var(--murmur-lamp, #e0a032);\n  --surface: #fbf8f1;\n  --raised: #f2ece0;\n  --ink: #211c17;\n  --muted: #675c4f;\n  --line: #e2d8c5;\n  --focus: #b4541d;\n  --danger: #a1301f;\n  --radius: var(--murmur-radius, 16px);\n  --shadow: 0 18px 48px -12px rgb(33 28 23 / 0.28), 0 4px 12px -4px rgb(33 28 23 / 0.16);\n  --ease: cubic-bezier(0.16, 1, 0.3, 1);\n  all: initial;\n  position: fixed;\n  z-index: var(--murmur-z, 2147483000);\n  inset: auto 20px 20px auto;\n  font-family: var(--murmur-font, inherit);\n  font-size: 15px;\n  line-height: 1.5;\n  color: var(--ink);\n  color-scheme: light;\n  -webkit-font-smoothing: antialiased;\n}\n:host([position=\"bottom-left\"]) { inset: auto auto 20px 20px; }\n:host([inline]) { position: relative; inset: auto; z-index: auto; display: block; height: 100%; }\n:host([theme=\"dark\"]) { --surface: #1c1814; --raised: #26211b; --ink: #f2ead9; --muted: #b8ab95; --line: #3b342b; --focus: #f0a35c; --danger: #f08a73; color-scheme: dark; }\n@media (prefers-color-scheme: dark) {\n  :host(:not([theme=\"light\"])) { --surface: #1c1814; --raised: #26211b; --ink: #f2ead9; --muted: #b8ab95; --line: #3b342b; --focus: #f0a35c; --danger: #f08a73; color-scheme: dark; }\n  :host(:not([theme=\"light\"])) { --murmur-accent-default: #f2ead9; }\n}\n@media print { :host { display: none !important; } }\n* { box-sizing: border-box; }\n[hidden] { display: none !important; }\n::selection { background: color-mix(in oklab, var(--lamp) 45%, transparent); color: inherit; }\nbutton, input, select, textarea { font: inherit; color: inherit; }\nbutton { cursor: pointer; }\n:focus-visible { outline: 2px solid var(--focus); outline-offset: 2px; }\na { color: inherit; text-underline-offset: 0.2em; text-decoration-thickness: 1px; }\n\n/* Launcher */\n.launcher {\n  display: inline-flex; align-items: center; gap: 10px;\n  height: 52px; padding: 0 20px 0 16px; border: 0; border-radius: 26px;\n  background: var(--accent); color: var(--accent-ink);\n  box-shadow: var(--shadow); font-weight: 600; letter-spacing: 0.01em;\n  transition: transform 220ms var(--ease), box-shadow 220ms var(--ease);\n}\n.launcher:hover { transform: translateY(-2px); }\n.launcher:active { transform: translateY(0); }\n.launcher[aria-expanded=\"true\"] { display: none; }\n.lamp {\n  width: 8px; height: 8px; border-radius: 50%; flex: none;\n  background: var(--lamp); box-shadow: 0 0 0 3px color-mix(in oklab, var(--lamp) 28%, transparent);\n}\n.lamp.off { background: transparent; box-shadow: inset 0 0 0 1.5px var(--muted); }\n.lamp.busy { animation: blink 1.1s steps(2, jump-none) infinite; }\n@keyframes blink { 50% { opacity: 0.25; } }\n\n/* Panel */\n.panel {\n  position: absolute; right: 0; bottom: 0;\n  display: flex; flex-direction: column;\n  width: min(400px, calc(100vw - 40px)); height: min(640px, calc(100dvh - 40px));\n  background: var(--surface); border-radius: var(--radius); box-shadow: var(--shadow);\n  overflow: hidden; transform-origin: bottom right;\n  animation: rise 320ms var(--ease);\n}\n:host([position=\"bottom-left\"]) .panel { right: auto; left: 0; transform-origin: bottom left; }\n:host([inline]) .panel { position: static; width: 100%; height: 100%; box-shadow: none; animation: none; border-radius: inherit; }\n@keyframes rise { from { opacity: 0; transform: translateY(12px) scale(0.98); } }\n@media (max-width: 480px) {\n  :host(:not([inline])) { inset: auto 12px 12px auto; }\n  :host(:not([inline])[position=\"bottom-left\"]) { inset: auto auto 12px 12px; }\n  :host(:not([inline])) .panel { position: fixed; inset: 0; width: 100%; height: 100dvh; border-radius: 0; }\n}\n\nheader {\n  display: flex; align-items: center; gap: 10px;\n  padding: 14px 10px 14px 18px; background: var(--accent); color: var(--accent-ink);\n}\nheader h2 { flex: 1; margin: 0; font-size: 16px; font-weight: 650; letter-spacing: 0.005em; line-height: 1.25; min-width: 0; overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }\n.icon-btn {\n  display: inline-grid; place-items: center; width: 36px; height: 36px; flex: none;\n  border: 0; border-radius: 10px; background: transparent; color: inherit;\n  transition: background 160ms;\n}\n.icon-btn:hover { background: color-mix(in oklab, currentColor 14%, transparent); }\nheader .icon-btn:focus-visible { outline-color: var(--accent-ink); }\n\n/* Conversation */\n.log {\n  flex: 1; overflow-y: auto; overscroll-behavior: contain;\n  padding: 18px 18px 8px; display: flex; flex-direction: column; gap: 16px;\n  scrollbar-width: thin; scrollbar-color: var(--line) transparent;\n}\n.msg { max-width: 100%; overflow-wrap: anywhere; }\n.msg p { margin: 0 0 0.6em; }\n.msg p:last-child { margin-bottom: 0; }\n.msg ul, .msg ol { margin: 0 0 0.6em; padding-left: 1.3em; }\n.msg li + li { margin-top: 0.2em; }\n.msg code { font-family: ui-monospace, SFMono-Regular, Menlo, Consolas, monospace; font-size: 0.88em; background: var(--raised); padding: 0.1em 0.35em; border-radius: 5px; }\n.msg pre { margin: 0 0 0.6em; padding: 10px 12px; background: var(--raised); border-radius: 10px; overflow-x: auto; }\n.msg pre code { background: none; padding: 0; }\n.user {\n  align-self: flex-end; max-width: 85%;\n  padding: 9px 14px; border-radius: 16px 16px 4px 16px;\n  background: var(--accent); color: var(--accent-ink); white-space: pre-wrap;\n}\n.bot { align-self: stretch; }\n.greeting { color: var(--muted); }\n\n.cite {\n  display: inline-grid; place-items: center; min-width: 1.45em; height: 1.45em; padding: 0 0.3em;\n  margin: 0 0.1em; vertical-align: 0.12em;\n  border-radius: 1em; box-shadow: inset 0 0 0 1.25px currentColor;\n  font-size: 0.72em; font-weight: 700; line-height: 1; text-decoration: none;\n  font-variant-numeric: tabular-nums; color: var(--muted);\n  transition: background 160ms, color 160ms;\n}\n.cite:hover, .cite:focus-visible { background: var(--lamp); color: #211c17; box-shadow: none; }\n\n.sources { margin: 12px 0 0; padding: 0; list-style: none; display: grid; gap: 2px; }\n.sources-label { margin: 18px 0 2px; font-size: 12px; font-weight: 650; letter-spacing: 0.06em; text-transform: uppercase; color: var(--muted); }\n.sources a {\n  display: grid; grid-template-columns: auto 1fr; align-items: baseline; gap: 10px;\n  padding: 7px 10px; margin: 0 -10px; border-radius: 10px; text-decoration: none;\n}\n.sources a:hover { background: var(--raised); }\n.sources .cite { margin: 0; }\n.sources .where { min-width: 0; font-size: 14px; line-height: 1.35; }\n.sources .where span { color: var(--muted); }\n\n.status { display: flex; align-items: center; gap: 10px; color: var(--muted); font-size: 14px; }\n.error { color: var(--danger); }\n.error p { margin: 0 0 8px; }\n.text-btn {\n  display: inline-flex; align-items: center; gap: 6px; padding: 6px 12px; border-radius: 9px;\n  border: 0; background: var(--raised); color: var(--ink); font-size: 14px; font-weight: 600;\n}\n.text-btn:hover { background: var(--line); }\n\n/* Composer */\nform.composer {\n  display: flex; align-items: flex-end; gap: 8px; margin: 8px 12px 0; padding: 6px 6px 6px 14px;\n  border-radius: 14px; background: var(--raised);\n  box-shadow: inset 0 0 0 1px var(--line); transition: box-shadow 160ms;\n}\nform.composer:focus-within { box-shadow: inset 0 0 0 1.5px var(--focus); }\ntextarea {\n  flex: 1; min-height: 38px; max-height: 140px; padding: 8px 0; resize: none;\n  border: 0; background: transparent; outline: none; line-height: 1.4;\n}\ntextarea:focus-visible { outline: none; }\ntextarea::placeholder, input::placeholder { color: var(--muted); opacity: 1; }\n.send {\n  display: inline-grid; place-items: center; width: 38px; height: 38px; flex: none;\n  border: 0; border-radius: 10px; background: var(--accent); color: var(--accent-ink);\n  transition: opacity 160ms;\n}\n.send:disabled { opacity: 0.35; cursor: default; }\nfooter { padding: 8px 18px 12px; font-size: 12px; color: var(--muted); text-align: center; }\nfooter a { text-decoration: none; }\nfooter a:hover { text-decoration: underline; }\n\n/* Settings */\n.setup { flex: 1; overflow-y: auto; padding: 18px; display: grid; gap: 16px; align-content: start; }\n.setup p.intro { margin: 0; color: var(--muted); font-size: 14px; }\n.field { display: grid; gap: 6px; }\n.field > span { font-size: 13px; font-weight: 650; }\n.field small { color: var(--muted); font-size: 12.5px; }\n.field input:not([type=\"checkbox\"]), .field select {\n  width: 100%; height: 40px; padding: 0 12px; border-radius: 10px;\n  border: 0; background: var(--raised); box-shadow: inset 0 0 0 1px var(--line);\n}\n.field input:focus-visible, .field select:focus-visible { outline: none; box-shadow: inset 0 0 0 1.5px var(--focus); }\n.check { display: flex; align-items: center; gap: 10px; font-size: 14px; }\n.check input { width: 18px; height: 18px; accent-color: var(--accent); margin: 0; }\n.presets { display: flex; gap: 8px; flex-wrap: wrap; }\n.row { display: flex; gap: 8px; align-items: center; flex-wrap: wrap; }\n.primary {\n  display: inline-flex; align-items: center; justify-content: center; gap: 8px;\n  height: 42px; padding: 0 18px; border: 0; border-radius: 11px;\n  background: var(--accent); color: var(--accent-ink); font-weight: 650;\n}\n.primary:disabled { opacity: 0.4; cursor: default; }\n.result { font-size: 14px; }\n.result.ok { color: inherit; }\n.result.ok::before { content: \"\"; display: inline-block; width: 8px; height: 8px; margin-right: 8px; border-radius: 50%; background: var(--lamp); }\n.result.fail { color: var(--danger); }\n.link-btn { padding: 0; border: 0; background: none; color: var(--muted); font-size: 13px; text-decoration: underline; text-underline-offset: 0.2em; justify-self: start; }\n\n.sr { position: absolute; width: 1px; height: 1px; overflow: hidden; clip-path: inset(50%); white-space: nowrap; }\n\n@media (prefers-reduced-motion: reduce) {\n  *, .panel { animation: none !important; transition: none !important; }\n}\n";
+export declare const STYLES = "\n:host {\n  --accent: var(--murmur-accent, #211c17);\n  --accent-ink: var(--murmur-accent-ink, #f6efe0);\n  --lamp: var(--murmur-lamp, #e0a032);\n  --surface: #fbf8f1;\n  --raised: #f2ece0;\n  --ink: #211c17;\n  --muted: #675c4f;\n  --line: #e2d8c5;\n  --focus: #b4541d;\n  --danger: #a1301f;\n  --radius: var(--murmur-radius, 16px);\n  --shadow: 0 18px 48px -12px rgb(33 28 23 / 0.28), 0 4px 12px -4px rgb(33 28 23 / 0.16);\n  --ease: cubic-bezier(0.16, 1, 0.3, 1);\n  all: initial;\n  position: fixed;\n  z-index: var(--murmur-z, 2147483000);\n  inset: auto 20px 20px auto;\n  font-family: var(--murmur-font, inherit);\n  font-size: 15px;\n  line-height: 1.5;\n  color: var(--ink);\n  color-scheme: light;\n  -webkit-font-smoothing: antialiased;\n}\n:host([position=\"bottom-left\"]) { inset: auto auto 20px 20px; }\n:host([inline]) { position: relative; inset: auto; z-index: auto; display: block; height: 100%; }\n:host([theme=\"dark\"]) { --surface: #1c1814; --raised: #26211b; --ink: #f2ead9; --muted: #b8ab95; --line: #3b342b; --focus: #f0a35c; --danger: #f08a73; color-scheme: dark; }\n@media (prefers-color-scheme: dark) {\n  :host(:not([theme=\"light\"])) { --surface: #1c1814; --raised: #26211b; --ink: #f2ead9; --muted: #b8ab95; --line: #3b342b; --focus: #f0a35c; --danger: #f08a73; color-scheme: dark; }\n  :host(:not([theme=\"light\"])) { --murmur-accent-default: #f2ead9; }\n}\n@media print { :host { display: none !important; } }\n* { box-sizing: border-box; }\n[hidden] { display: none !important; }\n::selection { background: color-mix(in oklab, var(--lamp) 45%, transparent); color: inherit; }\nbutton, input, select, textarea { font: inherit; color: inherit; }\nbutton { cursor: pointer; }\n:focus-visible { outline: 2px solid var(--focus); outline-offset: 2px; }\na { color: inherit; text-underline-offset: 0.2em; text-decoration-thickness: 1px; }\n\n/* Launcher */\n.launcher {\n  display: inline-flex; align-items: center; gap: 10px;\n  height: 52px; padding: 0 20px 0 16px; border: 0; border-radius: 26px;\n  background: var(--accent); color: var(--accent-ink);\n  box-shadow: var(--shadow); font-weight: 600; letter-spacing: 0.01em;\n  transition: transform 220ms var(--ease), box-shadow 220ms var(--ease);\n}\n.launcher:hover { transform: translateY(-2px); }\n.launcher:active { transform: translateY(0); }\n.launcher[aria-expanded=\"true\"] { display: none; }\n.lamp {\n  width: 8px; height: 8px; border-radius: 50%; flex: none;\n  background: var(--lamp); box-shadow: 0 0 0 3px color-mix(in oklab, var(--lamp) 28%, transparent);\n}\n.lamp.off { background: transparent; box-shadow: inset 0 0 0 1.5px var(--muted); }\n.lamp.busy { animation: blink 1.1s steps(2, jump-none) infinite; }\n@keyframes blink { 50% { opacity: 0.25; } }\n\n/* Panel */\n.panel {\n  position: absolute; right: 0; bottom: 0;\n  display: flex; flex-direction: column;\n  width: min(400px, calc(100vw - 40px)); height: min(640px, calc(100dvh - 40px));\n  background: var(--surface); border-radius: var(--radius); box-shadow: var(--shadow);\n  overflow: hidden; transform-origin: bottom right;\n  animation: rise 320ms var(--ease);\n}\n:host([position=\"bottom-left\"]) .panel { right: auto; left: 0; transform-origin: bottom left; }\n:host([inline]) .panel { position: static; width: 100%; height: 100%; box-shadow: none; animation: none; border-radius: inherit; }\n@keyframes rise { from { opacity: 0; transform: translateY(12px) scale(0.98); } }\n@media (max-width: 480px) {\n  :host(:not([inline])) { inset: auto 12px 12px auto; }\n  :host(:not([inline])[position=\"bottom-left\"]) { inset: auto auto 12px 12px; }\n  :host(:not([inline])) .panel { position: fixed; inset: 0; width: 100%; height: 100dvh; border-radius: 0; }\n}\n\nheader {\n  display: flex; align-items: center; gap: 10px;\n  padding: 14px 10px 14px 18px; background: var(--accent); color: var(--accent-ink);\n}\nheader h2 { flex: 1; margin: 0; font-size: 16px; font-weight: 650; letter-spacing: 0.005em; line-height: 1.25; min-width: 0; overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }\n.icon-btn {\n  display: inline-grid; place-items: center; width: 36px; height: 36px; flex: none;\n  border: 0; border-radius: 10px; background: transparent; color: inherit;\n  transition: background 160ms;\n}\n.icon-btn:hover { background: color-mix(in oklab, currentColor 14%, transparent); }\nheader .icon-btn:focus-visible { outline-color: var(--accent-ink); }\n\n/* Conversation */\n.log {\n  flex: 1; overflow-y: auto; overscroll-behavior: contain;\n  padding: 18px 18px 8px; display: flex; flex-direction: column; gap: 16px;\n  scrollbar-width: thin; scrollbar-color: var(--line) transparent;\n}\n.msg { max-width: 100%; overflow-wrap: anywhere; }\n.msg p { margin: 0 0 0.6em; }\n.msg p:last-child { margin-bottom: 0; }\n.msg ul, .msg ol { margin: 0 0 0.6em; padding-left: 1.3em; }\n.msg li + li { margin-top: 0.2em; }\n.msg code { font-family: ui-monospace, SFMono-Regular, Menlo, Consolas, monospace; font-size: 0.88em; background: var(--raised); padding: 0.1em 0.35em; border-radius: 5px; }\n.msg pre { margin: 0 0 0.6em; padding: 10px 12px; background: var(--raised); border-radius: 10px; overflow-x: auto; }\n.msg pre code { background: none; padding: 0; }\n.msg .table { margin: 0 0 0.6em; overflow-x: auto; border-radius: 10px; box-shadow: inset 0 0 0 1px var(--line); }\n.msg table { width: 100%; border-collapse: collapse; font-size: 14px; }\n.msg th, .msg td { padding: 6px 10px; text-align: left; vertical-align: top; border-bottom: 1px solid var(--line); }\n.msg tr:last-child td { border-bottom: 0; }\n.msg th { font-weight: 650; background: var(--raised); }\n.msg blockquote { margin: 0 0 0.6em; padding: 2px 0 2px 12px; color: var(--muted); box-shadow: inset 1px 0 0 var(--line); }\n.msg hr { margin: 0.8em 0; border: 0; border-top: 1px solid var(--line); }\n.user {\n  align-self: flex-end; max-width: 85%;\n  padding: 9px 14px; border-radius: 16px 16px 4px 16px;\n  background: var(--accent); color: var(--accent-ink); white-space: pre-wrap;\n}\n.bot { align-self: stretch; }\n.greeting { color: var(--muted); }\n\n.cite {\n  display: inline-grid; place-items: center; min-width: 1.45em; height: 1.45em; padding: 0 0.3em;\n  margin: 0 0.1em; vertical-align: 0.12em;\n  border-radius: 1em; box-shadow: inset 0 0 0 1.25px currentColor;\n  font-size: 0.72em; font-weight: 700; line-height: 1; text-decoration: none;\n  font-variant-numeric: tabular-nums; color: var(--muted);\n  transition: background 160ms, color 160ms;\n}\n.cite:hover, .cite:focus-visible { background: var(--lamp); color: #211c17; box-shadow: none; }\n\n.sources { margin: 12px 0 0; padding: 0; list-style: none; display: grid; gap: 2px; }\n.sources-label { margin: 18px 0 2px; font-size: 12px; font-weight: 650; letter-spacing: 0.06em; text-transform: uppercase; color: var(--muted); }\n.sources a {\n  display: grid; grid-template-columns: auto 1fr; align-items: baseline; gap: 10px;\n  padding: 7px 10px; margin: 0 -10px; border-radius: 10px; text-decoration: none;\n}\n.sources a:hover { background: var(--raised); }\n.sources .cite { margin: 0; }\n.sources .where { min-width: 0; font-size: 14px; line-height: 1.35; }\n.sources .where span { color: var(--muted); }\n\n.status { display: flex; align-items: center; gap: 10px; color: var(--muted); font-size: 14px; }\n.error { color: var(--danger); }\n.error p { margin: 0 0 8px; }\n.text-btn {\n  display: inline-flex; align-items: center; gap: 6px; padding: 6px 12px; border-radius: 9px;\n  border: 0; background: var(--raised); color: var(--ink); font-size: 14px; font-weight: 600;\n}\n.text-btn:hover { background: var(--line); }\n\n/* Composer */\nform.composer {\n  display: flex; align-items: flex-end; gap: 8px; margin: 8px 12px 0; padding: 6px 6px 6px 14px;\n  border-radius: 14px; background: var(--raised);\n  box-shadow: inset 0 0 0 1px var(--line); transition: box-shadow 160ms;\n}\nform.composer:focus-within { box-shadow: inset 0 0 0 1.5px var(--focus); }\ntextarea {\n  flex: 1; min-height: 38px; max-height: 140px; padding: 8px 0; resize: none;\n  border: 0; background: transparent; outline: none; line-height: 1.4;\n}\ntextarea:focus-visible { outline: none; }\ntextarea::placeholder, input::placeholder { color: var(--muted); opacity: 1; }\n.send {\n  display: inline-grid; place-items: center; width: 38px; height: 38px; flex: none;\n  border: 0; border-radius: 10px; background: var(--accent); color: var(--accent-ink);\n  transition: opacity 160ms;\n}\n.send:disabled { opacity: 0.35; cursor: default; }\nfooter { padding: 8px 18px 12px; font-size: 12px; color: var(--muted); text-align: center; }\nfooter a { text-decoration: none; }\nfooter a:hover { text-decoration: underline; }\n\n/* Settings */\n.setup { flex: 1; overflow-y: auto; padding: 18px; display: grid; gap: 16px; align-content: start; }\n.setup p.intro { margin: 0; color: var(--muted); font-size: 14px; }\n.field { display: grid; gap: 6px; }\n.field > span { font-size: 13px; font-weight: 650; }\n.field small { color: var(--muted); font-size: 12.5px; }\n.field input:not([type=\"checkbox\"]), .field select {\n  width: 100%; height: 40px; padding: 0 12px; border-radius: 10px;\n  border: 0; background: var(--raised); box-shadow: inset 0 0 0 1px var(--line);\n}\n.field input:focus-visible, .field select:focus-visible { outline: none; box-shadow: inset 0 0 0 1.5px var(--focus); }\n.check { display: flex; align-items: center; gap: 10px; font-size: 14px; }\n.check input { width: 18px; height: 18px; accent-color: var(--accent); margin: 0; }\n.presets { display: flex; gap: 8px; flex-wrap: wrap; }\n.row { display: flex; gap: 8px; align-items: center; flex-wrap: wrap; }\n.primary {\n  display: inline-flex; align-items: center; justify-content: center; gap: 8px;\n  height: 42px; padding: 0 18px; border: 0; border-radius: 11px;\n  background: var(--accent); color: var(--accent-ink); font-weight: 650;\n}\n.primary:disabled { opacity: 0.4; cursor: default; }\n.result { font-size: 14px; }\n.result.ok { color: inherit; }\n.result.ok::before { content: \"\"; display: inline-block; width: 8px; height: 8px; margin-right: 8px; border-radius: 50%; background: var(--lamp); }\n.result.fail { color: var(--danger); }\n.local-choice { display: grid; gap: 6px; padding: 14px; border-radius: 12px; background: var(--raised); }\n.local-choice small { color: var(--muted); font-size: 12.5px; }\n.local-choice .link-btn { margin-top: 2px; }\np.or { margin: 0; color: var(--muted); font-size: 13px; text-align: center; }\n.consent { display: grid; gap: 10px; padding: 14px; border-radius: 12px; background: var(--raised); }\n.consent p { margin: 0; }\n.status .grow { display: grid; gap: 6px; flex: 1; }\nprogress { width: 100%; height: 6px; appearance: none; border: 0; border-radius: 3px; background: var(--line); overflow: hidden; }\nprogress::-webkit-progress-bar { background: var(--line); }\nprogress::-webkit-progress-value { background: var(--lamp); }\nprogress::-moz-progress-bar { background: var(--lamp); }\n.link-btn { padding: 0; border: 0; background: none; color: var(--muted); font-size: 13px; text-decoration: underline; text-underline-offset: 0.2em; justify-self: start; }\n\n.sr { position: absolute; width: 1px; height: 1px; overflow: hidden; clip-path: inset(50%); white-space: nowrap; }\n\n@media (prefers-reduced-motion: reduce) {\n  *, .panel { animation: none !important; transition: none !important; }\n}\n";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "murmur-rag",
-  "version": "1.0.0",
+  "version": "1.2.0",
   "description": "Embeddable AI chat for static sites: one script tag, a mini RAG over your own pages, any LLM through llm-connect (your proxy or the visitor's own key). No backend.",
   "author": "maxgfr",
   "license": "MIT",
@@ -45,6 +45,10 @@
       "types": "./dist/types/llm/index.d.ts",
       "default": "./dist/lib/llm/index.js"
     },
+    "./local": {
+      "types": "./dist/types/local/index.d.ts",
+      "default": "./dist/lib/local/index.js"
+    },
     "./widget": {
       "types": "./dist/types/widget/index.d.ts",
       "default": "./dist/lib/widget/index.js"
@@ -103,7 +107,8 @@
     "linkedom": "^0.18.14",
     "llm-connect": "^2.5.0",
     "llm-models": "^1.14.0",
-    "minisearch": "^7.2.0"
+    "minisearch": "^7.2.0",
+    "@ai-sdk/provider": "^4.0.34"
   },
   "devDependencies": {
     "@axe-core/playwright": "^4.14.0",
@@ -113,6 +118,15 @@
     "@types/bun": "^1.4.4",
     "conventional-changelog-conventionalcommits": "^9",
     "semantic-release": "^25.0.9",
-    "typescript": "^7.0.2"
+    "typescript": "^7.0.2",
+    "@huggingface/transformers": "^4.3.2"
+  },
+  "peerDependencies": {
+    "@huggingface/transformers": "^4.3.2"
+  },
+  "peerDependenciesMeta": {
+    "@huggingface/transformers": {
+      "optional": true
+    }
   }
 }