nlm-memory 0.20.0 → 0.20.1

package/dist/llm/ollama-code-embedder.js

@@ -14,39 +14,35 @@
  * Vectors are L2-normalised before return so vec0 cosine retrieval is correct.
  */
 import { LLMUnreachableError } from "../ports/llm-client.js";
-const DEFAULT_CODE_EMBED_MODEL = "hf.co/awhiteside/CodeRankEmbed-Q8_0-GGUF";
-const CODE_RANK_QUERY_PREFIX = "Represent this query for searching relevant code: ";
-function l2Normalize(v) {
-    let sumSq = 0;
-    for (let i = 0; i < v.length; i++) {
-        const x = v[i] ?? 0;
-        sumSq += x * x;
+import { l2Normalize } from "./ollama-client.js";
+export const DEFAULT_CODE_EMBED_MODEL = "hf.co/awhiteside/CodeRankEmbed-Q8_0-GGUF";
+export const CODE_RANK_QUERY_PREFIX = "Represent this query for searching relevant code: ";
+/** Apply the code-embed prefix convention for `model`. CodeRankEmbed prefixes
+ *  queries and leaves documents raw; nomic-family models use the asymmetric
+ *  search_query/search_document prefixes. Shared by both transports so the two
+ *  code embedders produce identical inputs (and thus interchangeable vectors). */
+export function buildCodeEmbedPrompt(text, role, model) {
+    if (model.startsWith("nomic")) {
+        return role === "query" ? `search_query: ${text}` : `search_document: ${text}`;
     }
-    if (sumSq === 0)
-        return v;
-    const norm = Math.sqrt(sumSq);
-    const out = new Float32Array(v.length);
-    for (let i = 0; i < v.length; i++)
-        out[i] = (v[i] ?? 0) / norm;
-    return out;
+    return role === "query" ? `${CODE_RANK_QUERY_PREFIX}${text}` : text;
 }
 export class OllamaCodeEmbedder {
     baseUrl;
     model;
-    isNomic;
     constructor(opts = {}) {
         this.baseUrl = opts.baseUrl ?? (process.env["OLLAMA_HOST"] ?? "http://localhost:11434");
         this.model = opts.model ?? (process.env["NLM_CODE_EMBED_MODEL"] ?? DEFAULT_CODE_EMBED_MODEL);
-        this.isNomic = this.model.startsWith("nomic");
     }
-    async embed(text, role) {
-        const prompt = this.buildPrompt(text, role);
+    async embed(text, role, signal) {
+        const prompt = buildCodeEmbedPrompt(text, role, this.model);
         let res;
         try {
             res = await fetch(`${this.baseUrl}/api/embeddings`, {
                 method: "POST",
                 headers: { "Content-Type": "application/json" },
                 body: JSON.stringify({ model: this.model, prompt }),
+                ...(signal ? { signal } : {}),
             });
         }
         catch (e) {
@@ -63,11 +59,5 @@ export class OllamaCodeEmbedder {
         const vector = l2Normalize(raw);
         return { vector, dim: vector.length };
     }
-    buildPrompt(text, role) {
-        if (this.isNomic) {
-            return role === "query" ? `search_query: ${text}` : `search_document: ${text}`;
-        }
-        return role === "query" ? `${CODE_RANK_QUERY_PREFIX}${text}` : text;
-    }
 }
 //# sourceMappingURL=ollama-code-embedder.js.map

package/dist/llm/ollama-code-embedder.js.map

@@ -1 +1 @@
-{"version":3,"file":"ollama-code-embedder.js","sourceRoot":"","sources":["../../src/llm/ollama-code-embedder.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAGH,OAAO,EAAE,mBAAmB,EAAE,MAAM,sBAAsB,CAAC;AAE3D,MAAM,wBAAwB,GAAG,0CAA0C,CAAC;AAC5E,MAAM,sBAAsB,GAAG,oDAAoD,CAAC;AAEpF,SAAS,WAAW,CAAC,CAAe;IAClC,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAClC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;QACpB,KAAK,IAAI,CAAC,GAAG,CAAC,CAAC;IACjB,CAAC;IACD,IAAI,KAAK,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IAC1B,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;IAC9B,MAAM,GAAG,GAAG,IAAI,YAAY,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;IACvC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,EAAE;QAAE,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,IAAI,CAAC;IAC/D,OAAO,GAAG,CAAC;AACb,CAAC;AAOD,MAAM,OAAO,kBAAkB;IACZ,OAAO,CAAS;IAChB,KAAK,CAAS;IACd,OAAO,CAAU;IAElC,YAAY,OAAkC,EAAE;QAC9C,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,OAAO,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC,IAAI,wBAAwB,CAAC,CAAC;QACxF,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,sBAAsB,CAAC,IAAI,wBAAwB,CAAC,CAAC;QAC7F,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC;IAChD,CAAC;IAED,KAAK,CAAC,KAAK,CAAC,IAAY,EAAE,IAA0B;QAClD,MAAM,MAAM,GAAG,IAAI,CAAC,WAAW,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;QAC5C,IAAI,GAAa,CAAC;QAClB,IAAI,CAAC;YACH,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,IAAI,CAAC,OAAO,iBAAiB,EAAE;gBAClD,MAAM,EAAE,MAAM;gBACd,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE;gBAC/C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,MAAM,EAAE,CAAC;aACpD,CAAC,CAAC;QACL,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,MAAM,IAAI,mBAAmB,CAAC,sBAAsB,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;QACnE,CAAC;QACD,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;YACZ,MAAM,IAAI,mBAAmB,CAAC,sBAAsB,EAAE,QAAQ,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC;QAC9E,CAAC;QACD,MAAM,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAA6B,CAAC;QAC5D,IAAI,CAAC,IAAI,CAAC,SAAS,IAAI,IAAI,CAAC,SAAS,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACnD,MAAM,IAAI,mBAAmB,CAAC,sBAAsB,EAAE,iBAAiB,CAAC,CAAC;QAC3E,CAAC;QACD,MAAM,GAAG,GAAG,IAAI,YAAY,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QAC7C,MAAM,MAAM,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC;QAChC,OAAO,EAAE,MAAM,EAAE,GAAG,EAAE,MAAM,CAAC,MAAM,EAAE,CAAC;IACxC,CAAC;IAEO,WAAW,CAAC,IAAY,EAAE,IAA0B;QAC1D,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;YACjB,OAAO,IAAI,KAAK,OAAO,CAAC,CAAC,CAAC,iBAAiB,IAAI,EAAE,CAAC,CAAC,CAAC,oBAAoB,IAAI,EAAE,CAAC;QACjF,CAAC;QACD,OAAO,IAAI,KAAK,OAAO,CAAC,CAAC,CAAC,GAAG,sBAAsB,GAAG,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;IACtE,CAAC;CACF"}
+{"version":3,"file":"ollama-code-embedder.js","sourceRoot":"","sources":["../../src/llm/ollama-code-embedder.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAGH,OAAO,EAAE,mBAAmB,EAAE,MAAM,sBAAsB,CAAC;AAC3D,OAAO,EAAE,WAAW,EAAE,MAAM,oBAAoB,CAAC;AAEjD,MAAM,CAAC,MAAM,wBAAwB,GAAG,0CAA0C,CAAC;AACnF,MAAM,CAAC,MAAM,sBAAsB,GAAG,oDAAoD,CAAC;AAE3F;;;kFAGkF;AAClF,MAAM,UAAU,oBAAoB,CAAC,IAAY,EAAE,IAA0B,EAAE,KAAa;IAC1F,IAAI,KAAK,CAAC,UAAU,CAAC,OAAO,CAAC,EAAE,CAAC;QAC9B,OAAO,IAAI,KAAK,OAAO,CAAC,CAAC,CAAC,iBAAiB,IAAI,EAAE,CAAC,CAAC,CAAC,oBAAoB,IAAI,EAAE,CAAC;IACjF,CAAC;IACD,OAAO,IAAI,KAAK,OAAO,CAAC,CAAC,CAAC,GAAG,sBAAsB,GAAG,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;AACtE,CAAC;AAOD,MAAM,OAAO,kBAAkB;IACZ,OAAO,CAAS;IAChB,KAAK,CAAS;IAE/B,YAAY,OAAkC,EAAE;QAC9C,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,OAAO,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC,IAAI,wBAAwB,CAAC,CAAC;QACxF,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,sBAAsB,CAAC,IAAI,wBAAwB,CAAC,CAAC;IAC/F,CAAC;IAED,KAAK,CAAC,KAAK,CAAC,IAAY,EAAE,IAA0B,EAAE,MAAoB;QACxE,MAAM,MAAM,GAAG,oBAAoB,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC;QAC5D,IAAI,GAAa,CAAC;QAClB,IAAI,CAAC;YACH,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,IAAI,CAAC,OAAO,iBAAiB,EAAE;gBAClD,MAAM,EAAE,MAAM;gBACd,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE;gBAC/C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,MAAM,EAAE,CAAC;gBACnD,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAC9B,CAAC,CAAC;QACL,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,MAAM,IAAI,mBAAmB,CAAC,sBAAsB,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;QACnE,CAAC;QACD,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;YACZ,MAAM,IAAI,mBAAmB,CAAC,sBAAsB,EAAE,QAAQ,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC;QAC9E,CAAC;QACD,MAAM,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAA6B,CAAC;QAC5D,IAAI,CAAC,IAAI,CAAC,SAAS,IAAI,IAAI,CAAC,SAAS,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACnD,MAAM,IAAI,mBAAmB,CAAC,sBAAsB,EAAE,iBAAiB,CAAC,CAAC;QAC3E,CAAC;QACD,MAAM,GAAG,GAAG,IAAI,YAAY,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QAC7C,MAAM,MAAM,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC;QAChC,OAAO,EAAE,MAAM,EAAE,GAAG,EAAE,MAAM,CAAC,MAAM,EAAE,CAAC;IACxC,CAAC;CACF"}

package/dist/llm/openai-code-embedder-client.d.ts

@@ -0,0 +1,25 @@
+/**
+ * OpenAICodeEmbedderClient — the code-lane embedder over any OpenAI-compatible
+ * `/v1/embeddings` endpoint (LM Studio, oMLX, vLLM, cloud). It mirrors
+ * OllamaCodeEmbedder's contract exactly — same CodeRankEmbed/nomic prefix
+ * convention (via the shared `buildCodeEmbedPrompt`) and the same
+ * L2-normalization — so vectors are interchangeable with the Ollama transport.
+ * Only the request/response shape differs (`{model, input}` -> `data[0].embedding`).
+ */
+import type { CodeEmbedder, EmbedCodeResult } from "../ports/code-embedder.js";
+export type FetchImpl = typeof fetch;
+export interface OpenAICodeEmbedderClientOptions {
+    readonly baseUrl: string;
+    readonly model?: string;
+    /** Optional — local servers ignore it; a placeholder is sent when unset. */
+    readonly apiKey?: string;
+    readonly fetchImpl?: FetchImpl;
+}
+export declare class OpenAICodeEmbedderClient implements CodeEmbedder {
+    private readonly baseUrl;
+    private readonly model;
+    private readonly apiKey;
+    private readonly fetchImpl;
+    constructor(opts: OpenAICodeEmbedderClientOptions);
+    embed(text: string, role: "query" | "document", signal?: AbortSignal): Promise<EmbedCodeResult>;
+}

package/dist/llm/openai-code-embedder-client.js

@@ -0,0 +1,52 @@
+/**
+ * OpenAICodeEmbedderClient — the code-lane embedder over any OpenAI-compatible
+ * `/v1/embeddings` endpoint (LM Studio, oMLX, vLLM, cloud). It mirrors
+ * OllamaCodeEmbedder's contract exactly — same CodeRankEmbed/nomic prefix
+ * convention (via the shared `buildCodeEmbedPrompt`) and the same
+ * L2-normalization — so vectors are interchangeable with the Ollama transport.
+ * Only the request/response shape differs (`{model, input}` -> `data[0].embedding`).
+ */
+import { LLMUnreachableError } from "../ports/llm-client.js";
+import { l2Normalize } from "./ollama-client.js";
+import { DEFAULT_CODE_EMBED_MODEL, buildCodeEmbedPrompt } from "./ollama-code-embedder.js";
+export class OpenAICodeEmbedderClient {
+    baseUrl;
+    model;
+    apiKey;
+    fetchImpl;
+    constructor(opts) {
+        this.baseUrl = opts.baseUrl.replace(/\/+$/, "");
+        this.model = opts.model ?? DEFAULT_CODE_EMBED_MODEL;
+        this.apiKey = opts.apiKey ?? "local";
+        this.fetchImpl = opts.fetchImpl ?? fetch;
+    }
+    async embed(text, role, signal) {
+        const input = buildCodeEmbedPrompt(text, role, this.model);
+        let res;
+        try {
+            res = await this.fetchImpl(`${this.baseUrl}/embeddings`, {
+                method: "POST",
+                headers: {
+                    "Content-Type": "application/json",
+                    Authorization: `Bearer ${this.apiKey}`,
+                },
+                body: JSON.stringify({ model: this.model, input }),
+                ...(signal ? { signal } : {}),
+            });
+        }
+        catch (e) {
+            throw new LLMUnreachableError("openai-code-embedder", String(e));
+        }
+        if (!res.ok) {
+            throw new LLMUnreachableError("openai-code-embedder", `HTTP ${res.status}`);
+        }
+        const data = (await res.json());
+        const embedding = data.data?.[0]?.embedding;
+        if (!embedding || embedding.length === 0) {
+            throw new LLMUnreachableError("openai-code-embedder", "empty embedding");
+        }
+        const vector = l2Normalize(new Float32Array(embedding));
+        return { vector, dim: vector.length };
+    }
+}
+//# sourceMappingURL=openai-code-embedder-client.js.map

package/dist/llm/openai-code-embedder-client.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"openai-code-embedder-client.js","sourceRoot":"","sources":["../../src/llm/openai-code-embedder-client.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAGH,OAAO,EAAE,mBAAmB,EAAE,MAAM,sBAAsB,CAAC;AAC3D,OAAO,EAAE,WAAW,EAAE,MAAM,oBAAoB,CAAC;AACjD,OAAO,EAAE,wBAAwB,EAAE,oBAAoB,EAAE,MAAM,2BAA2B,CAAC;AAgB3F,MAAM,OAAO,wBAAwB;IAClB,OAAO,CAAS;IAChB,KAAK,CAAS;IACd,MAAM,CAAS;IACf,SAAS,CAAY;IAEtC,YAAY,IAAqC;QAC/C,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;QAChD,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,wBAAwB,CAAC;QACpD,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,MAAM,IAAI,OAAO,CAAC;QACrC,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,KAAK,CAAC;IAC3C,CAAC;IAED,KAAK,CAAC,KAAK,CAAC,IAAY,EAAE,IAA0B,EAAE,MAAoB;QACxE,MAAM,KAAK,GAAG,oBAAoB,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC;QAC3D,IAAI,GAAa,CAAC;QAClB,IAAI,CAAC;YACH,GAAG,GAAG,MAAM,IAAI,CAAC,SAAS,CAAC,GAAG,IAAI,CAAC,OAAO,aAAa,EAAE;gBACvD,MAAM,EAAE,MAAM;gBACd,OAAO,EAAE;oBACP,cAAc,EAAE,kBAAkB;oBAClC,aAAa,EAAE,UAAU,IAAI,CAAC,MAAM,EAAE;iBACvC;gBACD,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,KAAK,EAAE,CAAC;gBAClD,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAC9B,CAAC,CAAC;QACL,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,MAAM,IAAI,mBAAmB,CAAC,sBAAsB,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;QACnE,CAAC;QACD,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;YACZ,MAAM,IAAI,mBAAmB,CAAC,sBAAsB,EAAE,QAAQ,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC;QAC9E,CAAC;QACD,MAAM,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAAuB,CAAC;QACtD,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,SAAS,CAAC;QAC5C,IAAI,CAAC,SAAS,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACzC,MAAM,IAAI,mBAAmB,CAAC,sBAAsB,EAAE,iBAAiB,CAAC,CAAC;QAC3E,CAAC;QACD,MAAM,MAAM,GAAG,WAAW,CAAC,IAAI,YAAY,CAAC,SAAS,CAAC,CAAC,CAAC;QACxD,OAAO,EAAE,MAAM,EAAE,GAAG,EAAE,MAAM,CAAC,MAAM,EAAE,CAAC;IACxC,CAAC;CACF"}

package/dist/llm/openai-embedder-client.d.ts

@@ -0,0 +1,41 @@
+/**
+ * OpenAIEmbedderClient — an embedder backed by any OpenAI-compatible
+ * `/v1/embeddings` endpoint (LM Studio, oMLX, vLLM, or a cloud API). It mirrors
+ * OllamaClient's embedding contract exactly so vectors are interchangeable:
+ * the same nomic asymmetric-retrieval prefixes (`search_query:` /
+ * `search_document:`), the same MAX_EMBED_CHARS truncation, and the same
+ * L2-normalization. Only the transport differs (POST `/embeddings` with
+ * `{model, input}` and a `data[0].embedding` response shape).
+ *
+ * This client only embeds. `classify()` and `rewriteForRecall()` are not
+ * supported and throw LLMUnreachableError so a RecallService that holds this as
+ * its `llm` fails open to the raw query rather than crashing.
+ */
+import type { ClassifyResult, EmbedResult, EmbeddingKind, LLMClient, RewriteResult } from "../ports/llm-client.js";
+export type FetchImpl = typeof fetch;
+/** Default model for the openai embedder provider — nomic-embed-text v1.5 as
+ *  served by LM Studio (`/v1/embeddings`). Exported so the info descriptor
+ *  reports the same default the client actually uses. */
+export declare const DEFAULT_OPENAI_EMBED_MODEL = "text-embedding-nomic-embed-text-v1.5";
+export interface OpenAIEmbedderClientOptions {
+    readonly baseUrl: string;
+    readonly model?: string;
+    /** Optional — local servers ignore it; a placeholder is sent when unset. */
+    readonly apiKey?: string;
+    readonly timeoutMs?: number;
+    readonly fetchImpl?: FetchImpl;
+}
+export declare class OpenAIEmbedderClient implements LLMClient {
+    private readonly baseUrl;
+    private readonly model;
+    private readonly apiKey;
+    private readonly timeoutMs;
+    private readonly fetchImpl;
+    constructor(opts: OpenAIEmbedderClientOptions);
+    embed(text: string, kind: EmbeddingKind, opts?: {
+        signal?: AbortSignal;
+    }): Promise<EmbedResult>;
+    classify(_transcript: string): Promise<ClassifyResult>;
+    rewriteForRecall(_query: string): Promise<RewriteResult>;
+    nameWorkstream(): Promise<string | null>;
+}

package/dist/llm/openai-embedder-client.js

@@ -0,0 +1,83 @@
+/**
+ * OpenAIEmbedderClient — an embedder backed by any OpenAI-compatible
+ * `/v1/embeddings` endpoint (LM Studio, oMLX, vLLM, or a cloud API). It mirrors
+ * OllamaClient's embedding contract exactly so vectors are interchangeable:
+ * the same nomic asymmetric-retrieval prefixes (`search_query:` /
+ * `search_document:`), the same MAX_EMBED_CHARS truncation, and the same
+ * L2-normalization. Only the transport differs (POST `/embeddings` with
+ * `{model, input}` and a `data[0].embedding` response shape).
+ *
+ * This client only embeds. `classify()` and `rewriteForRecall()` are not
+ * supported and throw LLMUnreachableError so a RecallService that holds this as
+ * its `llm` fails open to the raw query rather than crashing.
+ */
+import { LLMUnreachableError } from "../ports/llm-client.js";
+import { EMBED_PREFIXES, MAX_EMBED_CHARS, l2Normalize } from "./ollama-client.js";
+/** Default model for the openai embedder provider — nomic-embed-text v1.5 as
+ *  served by LM Studio (`/v1/embeddings`). Exported so the info descriptor
+ *  reports the same default the client actually uses. */
+export const DEFAULT_OPENAI_EMBED_MODEL = "text-embedding-nomic-embed-text-v1.5";
+export class OpenAIEmbedderClient {
+    baseUrl;
+    model;
+    apiKey;
+    timeoutMs;
+    fetchImpl;
+    constructor(opts) {
+        this.baseUrl = opts.baseUrl.replace(/\/+$/, "");
+        this.model = opts.model ?? DEFAULT_OPENAI_EMBED_MODEL;
+        this.apiKey = opts.apiKey ?? "local";
+        this.timeoutMs = opts.timeoutMs ?? 30_000;
+        this.fetchImpl = opts.fetchImpl ?? fetch;
+    }
+    async embed(text, kind, opts) {
+        const input = `${EMBED_PREFIXES[kind]}${text.slice(0, MAX_EMBED_CHARS)}`;
+        const controller = new AbortController();
+        const timer = setTimeout(() => controller.abort(), this.timeoutMs);
+        const onExternal = opts?.signal != null ? () => { controller.abort(); } : null;
+        if (onExternal != null && opts?.signal != null) {
+            opts.signal.addEventListener("abort", onExternal, { once: true });
+        }
+        try {
+            const res = await this.fetchImpl(`${this.baseUrl}/embeddings`, {
+                method: "POST",
+                headers: {
+                    "Content-Type": "application/json",
+                    Authorization: `Bearer ${this.apiKey}`,
+                },
+                body: JSON.stringify({ model: this.model, input }),
+                signal: controller.signal,
+            });
+            if (!res.ok) {
+                throw new LLMUnreachableError("openai-embedder", `status ${res.status}`);
+            }
+            const data = (await res.json());
+            const embedding = data.data?.[0]?.embedding;
+            if (!embedding || embedding.length === 0) {
+                throw new LLMUnreachableError("openai-embedder", "empty embedding");
+            }
+            return { vector: l2Normalize(new Float32Array(embedding)), model: this.model };
+        }
+        catch (e) {
+            if (e instanceof LLMUnreachableError)
+                throw e;
+            throw new LLMUnreachableError("openai-embedder", e);
+        }
+        finally {
+            clearTimeout(timer);
+            if (onExternal != null && opts?.signal != null) {
+                opts.signal.removeEventListener("abort", onExternal);
+            }
+        }
+    }
+    async classify(_transcript) {
+        throw new LLMUnreachableError("openai-embedder", "classify not supported — OpenAIEmbedderClient is an embedder only");
+    }
+    async rewriteForRecall(_query) {
+        throw new LLMUnreachableError("openai-embedder", "rewriteForRecall not supported — OpenAIEmbedderClient is an embedder only");
+    }
+    nameWorkstream() {
+        throw new Error("OpenAIEmbedderClient does not support nameWorkstream");
+    }
+}
+//# sourceMappingURL=openai-embedder-client.js.map

package/dist/llm/openai-embedder-client.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"openai-embedder-client.js","sourceRoot":"","sources":["../../src/llm/openai-embedder-client.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AASH,OAAO,EAAE,mBAAmB,EAAE,MAAM,sBAAsB,CAAC;AAC3D,OAAO,EAAE,cAAc,EAAE,eAAe,EAAE,WAAW,EAAE,MAAM,oBAAoB,CAAC;AAIlF;;yDAEyD;AACzD,MAAM,CAAC,MAAM,0BAA0B,GAAG,sCAAsC,CAAC;AAejF,MAAM,OAAO,oBAAoB;IACd,OAAO,CAAS;IAChB,KAAK,CAAS;IACd,MAAM,CAAS;IACf,SAAS,CAAS;IAClB,SAAS,CAAY;IAEtC,YAAY,IAAiC;QAC3C,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;QAChD,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,0BAA0B,CAAC;QACtD,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,MAAM,IAAI,OAAO,CAAC;QACrC,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,MAAM,CAAC;QAC1C,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,KAAK,CAAC;IAC3C,CAAC;IAED,KAAK,CAAC,KAAK,CAAC,IAAY,EAAE,IAAmB,EAAE,IAA+B;QAC5E,MAAM,KAAK,GAAG,GAAG,cAAc,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,eAAe,CAAC,EAAE,CAAC;QACzE,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;QACzC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,IAAI,CAAC,SAAS,CAAC,CAAC;QACnE,MAAM,UAAU,GAAG,IAAI,EAAE,MAAM,IAAI,IAAI,CAAC,CAAC,CAAC,GAAS,EAAE,GAAG,UAAU,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;QACrF,IAAI,UAAU,IAAI,IAAI,IAAI,IAAI,EAAE,MAAM,IAAI,IAAI,EAAE,CAAC;YAC/C,IAAI,CAAC,MAAM,CAAC,gBAAgB,CAAC,OAAO,EAAE,UAAU,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;QACpE,CAAC;QACD,IAAI,CAAC;YACH,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,SAAS,CAAC,GAAG,IAAI,CAAC,OAAO,aAAa,EAAE;gBAC7D,MAAM,EAAE,MAAM;gBACd,OAAO,EAAE;oBACP,cAAc,EAAE,kBAAkB;oBAClC,aAAa,EAAE,UAAU,IAAI,CAAC,MAAM,EAAE;iBACvC;gBACD,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,KAAK,EAAE,CAAC;gBAClD,MAAM,EAAE,UAAU,CAAC,MAAM;aAC1B,CAAC,CAAC;YACH,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;gBACZ,MAAM,IAAI,mBAAmB,CAAC,iBAAiB,EAAE,UAAU,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC;YAC3E,CAAC;YACD,MAAM,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAAuB,CAAC;YACtD,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,SAAS,CAAC;YAC5C,IAAI,CAAC,SAAS,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBACzC,MAAM,IAAI,mBAAmB,CAAC,iBAAiB,EAAE,iBAAiB,CAAC,CAAC;YACtE,CAAC;YACD,OAAO,EAAE,MAAM,EAAE,WAAW,CAAC,IAAI,YAAY,CAAC,SAAS,CAAC,CAAC,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,CAAC;QACjF,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,IAAI,CAAC,YAAY,mBAAmB;gBAAE,MAAM,CAAC,CAAC;YAC9C,MAAM,IAAI,mBAAmB,CAAC,iBAAiB,EAAE,CAAC,CAAC,CAAC;QACtD,CAAC;gBAAS,CAAC;YACT,YAAY,CAAC,KAAK,CAAC,CAAC;YACpB,IAAI,UAAU,IAAI,IAAI,IAAI,IAAI,EAAE,MAAM,IAAI,IAAI,EAAE,CAAC;gBAC/C,IAAI,CAAC,MAAM,CAAC,mBAAmB,CAAC,OAAO,EAAE,UAAU,CAAC,CAAC;YACvD,CAAC;QACH,CAAC;IACH,CAAC;IAED,KAAK,CAAC,QAAQ,CAAC,WAAmB;QAChC,MAAM,IAAI,mBAAmB,CAC3B,iBAAiB,EACjB,mEAAmE,CACpE,CAAC;IACJ,CAAC;IAED,KAAK,CAAC,gBAAgB,CAAC,MAAc;QACnC,MAAM,IAAI,mBAAmB,CAC3B,iBAAiB,EACjB,2EAA2E,CAC5E,CAAC;IACJ,CAAC;IAED,cAAc;QACZ,MAAM,IAAI,KAAK,CAAC,sDAAsD,CAAC,CAAC;IAC1E,CAAC;CACF"}

package/dist/mcp/server.d.ts

@@ -14,6 +14,7 @@ import type { FactRecallService } from "../core/recall-facts/fact-recall-service
 import type { RecallService } from "../core/recall/recall-service.js";
 import type { FactStore } from "../ports/fact-store.js";
 import type { SessionStore } from "../ports/session-store.js";
+import { type BuildWorkDigestDeps } from "../core/work-digest/build-work-digest.js";
 import type { FactKind, RecallKindFilter, RecallMode } from "../shared/types.js";
 export interface McpDeps {
     readonly recall: RecallService;
@@ -26,6 +27,15 @@ export interface McpDeps {
     /** Optional code embedder for recall_code semantic search. */
     readonly codeEmbedder?: import("../ports/code-embedder.js").CodeEmbedder;
     readonly installScope?: string;
+    /** Wire to enable the work_summary tool (operator daily work digest). */
+    readonly workDigest?: BuildWorkDigestDeps;
+    /** Wire to enable recall_workstream. Mirrors RollupDeps + the store for idOrLabel resolution. */
+    readonly workstreams?: {
+        readonly store: import("../ports/workstream-store.js").WorkstreamStore;
+        readonly sessions: Pick<SessionStore, "listSessionIdsByWorkstreams">;
+        readonly facts: Pick<FactStore, "listBySessions">;
+        readonly exemplars: Pick<import("../ports/code-exemplar-store.js").CodeExemplarStore, "listBySessions">;
+    };
 }
 export interface ToolResult {
     content: {
@@ -41,8 +51,40 @@ export interface RecallToolInput {
     mode: RecallMode | undefined;
     limit: number | undefined;
     rewrite: boolean | undefined;
+    workstream: string | undefined;
 }
-export declare function recallSessionsHandler(deps: McpDeps, input: Partial<RecallToolInput>): Promise<ToolResult>;
+/**
+ * Normalize an MCP client's self-reported identity (from the initialize
+ * handshake's clientInfo) into a runtime attribution string for the recall
+ * query-log. Returns null when no usable name is present, so the log stays
+ * honest about unknown callers. Trim + lowercase keeps aggregation clean
+ * (claude-code, cursor, hermes) and matches the hook-side runtime convention.
+ */
+export declare function mcpRuntimeFromClient(client: {
+    name?: string;
+} | undefined): string | null;
+export declare function recallSessionsHandler(deps: McpDeps, input: Partial<RecallToolInput>, runtime?: string | null): Promise<ToolResult>;
+export declare function workSummaryHandler(deps: McpDeps, input: {
+    date?: string | undefined;
+}): Promise<ToolResult>;
+export declare function recallWorkstreamHandler(deps: McpDeps, input: {
+    idOrLabel?: string;
+}, runtime?: string | null): Promise<ToolResult>;
+export declare function renameWorkstreamHandler(deps: McpDeps, input: {
+    idOrLabel?: string;
+    label?: string;
+}): Promise<ToolResult>;
+export declare function retireWorkstreamHandler(deps: McpDeps, input: {
+    idOrLabel?: string;
+}): Promise<ToolResult>;
+export declare function mergeWorkstreamsHandler(deps: McpDeps, input: {
+    from?: string;
+    into?: string;
+}): Promise<ToolResult>;
+export declare function rebindSessionHandler(deps: McpDeps, input: {
+    sessionId?: string;
+    workstream?: string;
+}): Promise<ToolResult>;
 export declare function getSessionHandler(deps: McpDeps, input: {
     id: string;
 }): Promise<ToolResult>;
@@ -56,7 +98,7 @@ export interface RecallFactsInput {
     minConfidence: number | undefined;
     limit: number | undefined;
 }
-export declare function recallFactsHandler(deps: McpDeps, input: Partial<RecallFactsInput>): Promise<ToolResult>;
+export declare function recallFactsHandler(deps: McpDeps, input: Partial<RecallFactsInput>, runtime?: string | null): Promise<ToolResult>;
 export declare function getFactHistoryHandler(deps: McpDeps, input: {
     subject: string;
     predicate: string | undefined;
@@ -78,4 +120,7 @@ export interface SupersedeFactInput {
 }
 export declare function supersedeFactHandler(deps: McpDeps, input: SupersedeFactInput): Promise<ToolResult>;
 export declare function citeSessionHandler(input: CiteSessionInput): Promise<ToolResult>;
+export declare function listMergeSuggestionsHandler(deps: McpDeps, input: {
+    minScore: number | undefined;
+}): Promise<ToolResult>;
 export declare function createMcpServer(deps: McpDeps): McpServer;