sverklo 0.2.12 → 0.2.14

package/dist/src/indexer/embedding-providers.js

@@ -0,0 +1,215 @@
+// Pluggable embedding providers (issue #9).
+//
+// Sverklo historically hardcoded the bundled all-MiniLM-L6-v2 ONNX
+// model. That's a great Pareto choice for the zero-config path but
+// leaves two legitimate user groups out:
+//
+//   1. Enterprise users with existing embedding infrastructure who
+//      want sverklo's index to share a similarity space with their
+//      RAG pipeline (Voyage AI, OpenAI, Cohere).
+//   2. Privacy-sensitive users who can't ship ONNX binaries and need
+//      to point at a local Ollama / llamafile endpoint.
+//
+// This module defines the provider interface and a registry. The
+// bundled ONNX model implements the interface as the "default"
+// provider and is still selected when the user sets nothing. All
+// other providers are additive — shipping a new one is a matter of
+// adding a class and registering it.
+//
+// Critical constraint: changing providers changes the embedding
+// dimension and the similarity space. We don't support mixing vectors
+// from different providers in the same index. The caller (Indexer)
+// checks the stored provider/dimensions against the current config
+// on startup and triggers a full rebuild if they don't match.
+// ────────────────────────────────────────────────────────────────────
+// Provider: default (bundled ONNX all-MiniLM-L6-v2)
+// ────────────────────────────────────────────────────────────────────
+import { embed as legacyEmbed, initEmbedder } from "./embedder.js";
+class BundledOnnxProvider {
+    name = "default";
+    dimensions = 384;
+    async init() {
+        await initEmbedder();
+    }
+    async embed(texts) {
+        return legacyEmbed(texts);
+    }
+}
+class OpenAIProvider {
+    name;
+    dimensions;
+    apiKey;
+    model;
+    endpoint;
+    constructor(opts) {
+        this.apiKey = opts.apiKey;
+        this.model = opts.model || "text-embedding-3-small";
+        // Default to 1536 for 3-small. Users targeting a different model
+        // MUST set SVERKLO_OPENAI_DIMENSIONS to match — we don't auto-probe
+        // because that would fire a billed request just to learn the size.
+        this.dimensions = opts.dimensions || 1536;
+        this.endpoint = opts.endpoint || "https://api.openai.com/v1/embeddings";
+        this.name = `openai:${this.model}`;
+    }
+    async init() {
+        // Smoke-test the endpoint with an empty ping. On failure, throw —
+        // the indexer wraps this in a try/catch and falls back to the
+        // bundled provider with a warning.
+        if (!this.apiKey) {
+            throw new Error("OpenAI embedding provider selected but SVERKLO_OPENAI_API_KEY is unset.");
+        }
+    }
+    async embed(texts) {
+        if (texts.length === 0)
+            return [];
+        // OpenAI's embeddings endpoint accepts a batch in a single call.
+        // We keep batches at <= 100 inputs to stay well under the 300k
+        // token request limit. The indexer already chunks at ~400 tokens
+        // per chunk so 100 × 400 = 40k tokens is safely under.
+        const out = [];
+        const BATCH = 100;
+        for (let i = 0; i < texts.length; i += BATCH) {
+            const batch = texts.slice(i, i + BATCH);
+            const res = await fetch(this.endpoint, {
+                method: "POST",
+                headers: {
+                    "content-type": "application/json",
+                    authorization: `Bearer ${this.apiKey}`,
+                },
+                body: JSON.stringify({
+                    model: this.model,
+                    input: batch,
+                    dimensions: this.dimensions,
+                }),
+            });
+            if (!res.ok) {
+                const body = await res.text().catch(() => "<no body>");
+                throw new Error(`OpenAI embeddings failed: ${res.status} ${res.statusText} — ${body.slice(0, 200)}`);
+            }
+            const json = (await res.json());
+            // Preserve input order — OpenAI is supposed to echo back sorted
+            // by index but we're defensive.
+            const sorted = [...json.data].sort((a, b) => a.index - b.index);
+            for (const row of sorted) {
+                out.push(new Float32Array(row.embedding));
+            }
+        }
+        return out;
+    }
+}
+class OllamaProvider {
+    name;
+    dimensions;
+    endpoint;
+    model;
+    constructor(opts) {
+        this.endpoint = opts.endpoint || "http://localhost:11434/api/embeddings";
+        this.model = opts.model || "nomic-embed-text";
+        // Nomic embed text is 768 dims. Users swapping models must set
+        // SVERKLO_OLLAMA_DIMENSIONS to match.
+        this.dimensions = opts.dimensions || 768;
+        this.name = `ollama:${this.model}`;
+    }
+    async init() {
+        // Probe with the read-only /api/tags listing endpoint instead of
+        // firing a real embedding request. Derives the tags endpoint from
+        // the embeddings endpoint so custom SVERKLO_OLLAMA_URL overrides
+        // still work. Failure falls back to the bundled provider via the
+        // factory wrapper.
+        const tagsEndpoint = this.endpoint.replace(/\/api\/embeddings\/?$/, "/api/tags");
+        try {
+            const res = await fetch(tagsEndpoint, { method: "GET" });
+            if (!res.ok) {
+                throw new Error(`Ollama probe failed: ${res.status} ${res.statusText}`);
+            }
+        }
+        catch (err) {
+            throw new Error(`Ollama embedding provider could not reach ${tagsEndpoint}. ` +
+                `Is Ollama running? Original error: ${err.message}`);
+        }
+    }
+    async embed(texts) {
+        // Ollama's embeddings endpoint takes one prompt per request, so
+        // we parallelize with a bounded concurrency of 4.
+        const out = new Array(texts.length);
+        const CONCURRENCY = 4;
+        let next = 0;
+        async function worker() {
+            while (true) {
+                const i = next++;
+                if (i >= texts.length)
+                    return;
+                const res = await fetch(this.endpoint, {
+                    method: "POST",
+                    headers: { "content-type": "application/json" },
+                    body: JSON.stringify({ model: this.model, prompt: texts[i] }),
+                });
+                if (!res.ok) {
+                    throw new Error(`Ollama embeddings failed on index ${i}: ${res.status}`);
+                }
+                const json = (await res.json());
+                out[i] = new Float32Array(json.embedding);
+            }
+        }
+        const workers = Array.from({ length: Math.min(CONCURRENCY, texts.length) }, () => worker.call(this));
+        await Promise.all(workers);
+        return out;
+    }
+}
+// ────────────────────────────────────────────────────────────────────
+// Factory
+// ────────────────────────────────────────────────────────────────────
+//
+// Reads config (env vars for now; .sverklo/config.json later) and
+// returns the selected provider. If init() throws, falls back to the
+// bundled default with a warning.
+import { log } from "../utils/logger.js";
+export async function createEmbeddingProvider(env = process.env) {
+    const providerName = (env.SVERKLO_EMBEDDING_PROVIDER || "default").toLowerCase();
+    let provider;
+    try {
+        switch (providerName) {
+            case "default":
+            case "bundled":
+            case "onnx":
+                provider = new BundledOnnxProvider();
+                break;
+            case "openai":
+                provider = new OpenAIProvider({
+                    apiKey: env.SVERKLO_OPENAI_API_KEY || "",
+                    model: env.SVERKLO_OPENAI_MODEL,
+                    dimensions: env.SVERKLO_OPENAI_DIMENSIONS
+                        ? parseInt(env.SVERKLO_OPENAI_DIMENSIONS, 10)
+                        : undefined,
+                });
+                break;
+            case "ollama":
+                provider = new OllamaProvider({
+                    endpoint: env.SVERKLO_OLLAMA_URL,
+                    model: env.SVERKLO_OLLAMA_MODEL,
+                    dimensions: env.SVERKLO_OLLAMA_DIMENSIONS
+                        ? parseInt(env.SVERKLO_OLLAMA_DIMENSIONS, 10)
+                        : undefined,
+                });
+                break;
+            default:
+                log(`[embedding] Unknown provider '${providerName}'. Falling back to default (bundled ONNX).`);
+                provider = new BundledOnnxProvider();
+        }
+        await provider.init();
+        if (providerName !== "default") {
+            log(`[embedding] Using ${provider.name} (${provider.dimensions} dims) — override with SVERKLO_EMBEDDING_PROVIDER.`);
+        }
+        return provider;
+    }
+    catch (err) {
+        log(`[embedding] Provider '${providerName}' init failed: ${err.message}. Falling back to default.`);
+        const fallback = new BundledOnnxProvider();
+        await fallback.init();
+        return fallback;
+    }
+}
+export function fingerprintOf(p) {
+    return { provider: p.name, dimensions: p.dimensions };
+}
+//# sourceMappingURL=embedding-providers.js.map

package/dist/src/indexer/embedding-providers.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"embedding-providers.js","sourceRoot":"","sources":["../../../src/indexer/embedding-providers.ts"],"names":[],"mappings":"AAAA,4CAA4C;AAC5C,EAAE;AACF,mEAAmE;AACnE,mEAAmE;AACnE,yCAAyC;AACzC,EAAE;AACF,mEAAmE;AACnE,mEAAmE;AACnE,iDAAiD;AACjD,qEAAqE;AACrE,wDAAwD;AACxD,EAAE;AACF,iEAAiE;AACjE,+DAA+D;AAC/D,iEAAiE;AACjE,mEAAmE;AACnE,qCAAqC;AACrC,EAAE;AACF,gEAAgE;AAChE,sEAAsE;AACtE,mEAAmE;AACnE,mEAAmE;AACnE,8DAA8D;AAkC9D,uEAAuE;AACvE,oDAAoD;AACpD,uEAAuE;AAEvE,OAAO,EAAE,KAAK,IAAI,WAAW,EAAE,YAAY,EAAE,MAAM,eAAe,CAAC;AAEnE,MAAM,mBAAmB;IACd,IAAI,GAAG,SAAS,CAAC;IACjB,UAAU,GAAG,GAAG,CAAC;IAE1B,KAAK,CAAC,IAAI;QACR,MAAM,YAAY,EAAE,CAAC;IACvB,CAAC;IAED,KAAK,CAAC,KAAK,CAAC,KAAe;QACzB,OAAO,WAAW,CAAC,KAAK,CAAC,CAAC;IAC5B,CAAC;CACF;AAeD,MAAM,cAAc;IACT,IAAI,CAAS;IACb,UAAU,CAAS;IACpB,MAAM,CAAS;IACf,KAAK,CAAS;IACd,QAAQ,CAAS;IAEzB,YAAY,IAAgF;QAC1F,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC;QAC1B,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,wBAAwB,CAAC;QACpD,iEAAiE;QACjE,oEAAoE;QACpE,mEAAmE;QACnE,IAAI,CAAC,UAAU,GAAG,IAAI,CAAC,UAAU,IAAI,IAAI,CAAC;QAC1C,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,QAAQ,IAAI,sCAAsC,CAAC;QACxE,IAAI,CAAC,IAAI,GAAG,UAAU,IAAI,CAAC,KAAK,EAAE,CAAC;IACrC,CAAC;IAED,KAAK,CAAC,IAAI;QACR,kEAAkE;QAClE,8DAA8D;QAC9D,mCAAmC;QACnC,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC;YACjB,MAAM,IAAI,KAAK,CACb,yEAAyE,CAC1E,CAAC;QACJ,CAAC;IACH,CAAC;IAED,KAAK,CAAC,KAAK,CAAC,KAAe;QACzB,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,EAAE,CAAC;QAElC,iEAAiE;QACjE,+DAA+D;QAC/D,iEAAiE;QACjE,uDAAuD;QACvD,MAAM,GAAG,GAAmB,EAAE,CAAC;QAC/B,MAAM,KAAK,GAAG,GAAG,CAAC;QAElB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,IAAI,KAAK,EAAE,CAAC;YAC7C,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,CAAC;YACxC,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,IAAI,CAAC,QAAQ,EAAE;gBACrC,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;oBACnB,KAAK,EAAE,IAAI,CAAC,KAAK;oBACjB,KAAK,EAAE,KAAK;oBACZ,UAAU,EAAE,IAAI,CAAC,UAAU;iBAC5B,CAAC;aACH,CAAC,CAAC;YACH,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;gBACZ,MAAM,IAAI,GAAG,MAAM,GAAG,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,WAAW,CAAC,CAAC;gBACvD,MAAM,IAAI,KAAK,CACb,6BAA6B,GAAG,CAAC,MAAM,IAAI,GAAG,CAAC,UAAU,MAAM,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CACpF,CAAC;YACJ,CAAC;YACD,MAAM,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAA4B,CAAC;YAC3D,gEAAgE;YAChE,gCAAgC;YAChC,MAAM,MAAM,GAAG,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC;YAChE,KAAK,MAAM,GAAG,IAAI,MAAM,EAAE,CAAC;gBACzB,GAAG,CAAC,IAAI,CAAC,IAAI,YAAY,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC;YAC5C,CAAC;QACH,CAAC;QAED,OAAO,GAAG,CAAC;IACb,CAAC;CACF;AAeD,MAAM,cAAc;IACT,IAAI,CAAS;IACb,UAAU,CAAS;IACpB,QAAQ,CAAS;IACjB,KAAK,CAAS;IAEtB,YAAY,IAAgE;QAC1E,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,QAAQ,IAAI,uCAAuC,CAAC;QACzE,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,kBAAkB,CAAC;QAC9C,+DAA+D;QAC/D,sCAAsC;QACtC,IAAI,CAAC,UAAU,GAAG,IAAI,CAAC,UAAU,IAAI,GAAG,CAAC;QACzC,IAAI,CAAC,IAAI,GAAG,UAAU,IAAI,CAAC,KAAK,EAAE,CAAC;IACrC,CAAC;IAED,KAAK,CAAC,IAAI;QACR,iEAAiE;QACjE,kEAAkE;QAClE,iEAAiE;QACjE,iEAAiE;QACjE,mBAAmB;QACnB,MAAM,YAAY,GAAG,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,uBAAuB,EAAE,WAAW,CAAC,CAAC;QACjF,IAAI,CAAC;YACH,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,YAAY,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC,CAAC;YACzD,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;gBACZ,MAAM,IAAI,KAAK,CAAC,wBAAwB,GAAG,CAAC,MAAM,IAAI,GAAG,CAAC,UAAU,EAAE,CAAC,CAAC;YAC1E,CAAC;QACH,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,MAAM,IAAI,KAAK,CACb,6CAA6C,YAAY,IAAI;gBAC3D,sCAAuC,GAAa,CAAC,OAAO,EAAE,CACjE,CAAC;QACJ,CAAC;IACH,CAAC;IAED,KAAK,CAAC,KAAK,CAAC,KAAe;QACzB,gEAAgE;QAChE,kDAAkD;QAClD,MAAM,GAAG,GAAmB,IAAI,KAAK,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;QACpD,MAAM,WAAW,GAAG,CAAC,CAAC;QACtB,IAAI,IAAI,GAAG,CAAC,CAAC;QAEb,KAAK,UAAU,MAAM;YACnB,OAAO,IAAI,EAAE,CAAC;gBACZ,MAAM,CAAC,GAAG,IAAI,EAAE,CAAC;gBACjB,IAAI,CAAC,IAAI,KAAK,CAAC,MAAM;oBAAE,OAAO;gBAC9B,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,IAAI,CAAC,QAAQ,EAAE;oBACrC,MAAM,EAAE,MAAM;oBACd,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE;oBAC/C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,MAAM,EAAE,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC;iBAC9D,CAAC,CAAC;gBACH,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;oBACZ,MAAM,IAAI,KAAK,CAAC,qCAAqC,CAAC,KAAK,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC;gBAC3E,CAAC;gBACD,MAAM,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAA4B,CAAC;gBAC3D,GAAG,CAAC,CAAC,CAAC,GAAG,IAAI,YAAY,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;YAC5C,CAAC;QACH,CAAC;QAED,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,WAAW,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,EAAE,GAAG,EAAE,CAC/E,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAClB,CAAC;QACF,MAAM,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;QAC3B,OAAO,GAAG,CAAC;IACb,CAAC;CACF;AAED,uEAAuE;AACvE,UAAU;AACV,uEAAuE;AACvE,EAAE;AACF,kEAAkE;AAClE,qEAAqE;AACrE,kCAAkC;AAElC,OAAO,EAAE,GAAG,EAAE,MAAM,oBAAoB,CAAC;AAEzC,MAAM,CAAC,KAAK,UAAU,uBAAuB,CAC3C,MAAyB,OAAO,CAAC,GAAG;IAEpC,MAAM,YAAY,GAAG,CAAC,GAAG,CAAC,0BAA0B,IAAI,SAAS,CAAC,CAAC,WAAW,EAAE,CAAC;IAEjF,IAAI,QAA2B,CAAC;IAEhC,IAAI,CAAC;QACH,QAAQ,YAAY,EAAE,CAAC;YACrB,KAAK,SAAS,CAAC;YACf,KAAK,SAAS,CAAC;YACf,KAAK,MAAM;gBACT,QAAQ,GAAG,IAAI,mBAAmB,EAAE,CAAC;gBACrC,MAAM;YAER,KAAK,QAAQ;gBACX,QAAQ,GAAG,IAAI,cAAc,CAAC;oBAC5B,MAAM,EAAE,GAAG,CAAC,sBAAsB,IAAI,EAAE;oBACxC,KAAK,EAAE,GAAG,CAAC,oBAAoB;oBAC/B,UAAU,EAAE,GAAG,CAAC,yBAAyB;wBACvC,CAAC,CAAC,QAAQ,CAAC,GAAG,CAAC,yBAAyB,EAAE,EAAE,CAAC;wBAC7C,CAAC,CAAC,SAAS;iBACd,CAAC,CAAC;gBACH,MAAM;YAER,KAAK,QAAQ;gBACX,QAAQ,GAAG,IAAI,cAAc,CAAC;oBAC5B,QAAQ,EAAE,GAAG,CAAC,kBAAkB;oBAChC,KAAK,EAAE,GAAG,CAAC,oBAAoB;oBAC/B,UAAU,EAAE,GAAG,CAAC,yBAAyB;wBACvC,CAAC,CAAC,QAAQ,CAAC,GAAG,CAAC,yBAAyB,EAAE,EAAE,CAAC;wBAC7C,CAAC,CAAC,SAAS;iBACd,CAAC,CAAC;gBACH,MAAM;YAER;gBACE,GAAG,CACD,iCAAiC,YAAY,4CAA4C,CAC1F,CAAC;gBACF,QAAQ,GAAG,IAAI,mBAAmB,EAAE,CAAC;QACzC,CAAC;QAED,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;QACtB,IAAI,YAAY,KAAK,SAAS,EAAE,CAAC;YAC/B,GAAG,CACD,qBAAqB,QAAQ,CAAC,IAAI,KAAK,QAAQ,CAAC,UAAU,oDAAoD,CAC/G,CAAC;QACJ,CAAC;QACD,OAAO,QAAQ,CAAC;IAClB,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,GAAG,CACD,yBAAyB,YAAY,kBAAmB,GAAa,CAAC,OAAO,4BAA4B,CAC1G,CAAC;QACF,MAAM,QAAQ,GAAG,IAAI,mBAAmB,EAAE,CAAC;QAC3C,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;QACtB,OAAO,QAAQ,CAAC;IAClB,CAAC;AACH,CAAC;AASD,MAAM,UAAU,aAAa,CAAC,CAAoB;IAChD,OAAO,EAAE,QAAQ,EAAE,CAAC,CAAC,IAAI,EAAE,UAAU,EAAE,CAAC,CAAC,UAAU,EAAE,CAAC;AACxD,CAAC"}

package/dist/src/indexer/embedding-providers.test.d.ts

@@ -0,0 +1 @@
+export {};

package/dist/src/indexer/embedding-providers.test.js

@@ -0,0 +1,83 @@
+import { describe, it, expect, beforeEach, afterEach, vi } from "vitest";
+import { createEmbeddingProvider, fingerprintOf } from "./embedding-providers.js";
+// We mock the underlying embedder module so tests don't try to load
+// the real ONNX runtime or download the model.
+vi.mock("./embedder.js", () => ({
+    initEmbedder: vi.fn(async () => { }),
+    embed: vi.fn(async (texts) => texts.map(() => new Float32Array(384))),
+}));
+describe("createEmbeddingProvider", () => {
+    const originalFetch = global.fetch;
+    beforeEach(() => {
+        vi.clearAllMocks();
+    });
+    afterEach(() => {
+        global.fetch = originalFetch;
+    });
+    it("defaults to the bundled ONNX provider when no env var is set", async () => {
+        const p = await createEmbeddingProvider({});
+        expect(p.name).toBe("default");
+        expect(p.dimensions).toBe(384);
+    });
+    it("accepts provider aliases (bundled, onnx)", async () => {
+        const p1 = await createEmbeddingProvider({ SVERKLO_EMBEDDING_PROVIDER: "bundled" });
+        const p2 = await createEmbeddingProvider({ SVERKLO_EMBEDDING_PROVIDER: "onnx" });
+        expect(p1.name).toBe("default");
+        expect(p2.name).toBe("default");
+    });
+    it("creates an OpenAI provider when requested and API key is set", async () => {
+        const p = await createEmbeddingProvider({
+            SVERKLO_EMBEDDING_PROVIDER: "openai",
+            SVERKLO_OPENAI_API_KEY: "sk-test",
+        });
+        expect(p.name).toContain("openai");
+        expect(p.dimensions).toBe(1536);
+    });
+    it("falls back to default when OpenAI is requested without an API key", async () => {
+        const p = await createEmbeddingProvider({
+            SVERKLO_EMBEDDING_PROVIDER: "openai",
+        });
+        // Init throws on missing key → factory falls back to default.
+        expect(p.name).toBe("default");
+    });
+    it("respects SVERKLO_OPENAI_DIMENSIONS override", async () => {
+        const p = await createEmbeddingProvider({
+            SVERKLO_EMBEDDING_PROVIDER: "openai",
+            SVERKLO_OPENAI_API_KEY: "sk-test",
+            SVERKLO_OPENAI_DIMENSIONS: "512",
+        });
+        expect(p.dimensions).toBe(512);
+    });
+    it("creates an Ollama provider when the endpoint probe succeeds", async () => {
+        global.fetch = vi.fn(async () => new Response(JSON.stringify({ embedding: new Array(768).fill(0) }), { status: 200 }));
+        const p = await createEmbeddingProvider({
+            SVERKLO_EMBEDDING_PROVIDER: "ollama",
+        });
+        expect(p.name).toContain("ollama");
+        expect(p.dimensions).toBe(768);
+    });
+    it("falls back to default when Ollama is unreachable", async () => {
+        global.fetch = vi.fn(async () => {
+            throw new Error("ECONNREFUSED");
+        });
+        const p = await createEmbeddingProvider({
+            SVERKLO_EMBEDDING_PROVIDER: "ollama",
+        });
+        expect(p.name).toBe("default");
+    });
+    it("falls back to default for unknown provider names", async () => {
+        const p = await createEmbeddingProvider({
+            SVERKLO_EMBEDDING_PROVIDER: "magic-ai",
+        });
+        expect(p.name).toBe("default");
+    });
+});
+describe("fingerprintOf", () => {
+    it("captures provider name and dimensions", async () => {
+        const p = await createEmbeddingProvider({});
+        const fp = fingerprintOf(p);
+        expect(fp.provider).toBe("default");
+        expect(fp.dimensions).toBe(384);
+    });
+});
+//# sourceMappingURL=embedding-providers.test.js.map

package/dist/src/indexer/embedding-providers.test.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"embedding-providers.test.js","sourceRoot":"","sources":["../../../src/indexer/embedding-providers.test.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,UAAU,EAAE,SAAS,EAAE,EAAE,EAAE,MAAM,QAAQ,CAAC;AACzE,OAAO,EAAE,uBAAuB,EAAE,aAAa,EAAE,MAAM,0BAA0B,CAAC;AAElF,oEAAoE;AACpE,+CAA+C;AAC/C,EAAE,CAAC,IAAI,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,CAAC;IAC9B,YAAY,EAAE,EAAE,CAAC,EAAE,CAAC,KAAK,IAAI,EAAE,GAAE,CAAC,CAAC;IACnC,KAAK,EAAE,EAAE,CAAC,EAAE,CAAC,KAAK,EAAE,KAAe,EAAE,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,IAAI,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC;CAChF,CAAC,CAAC,CAAC;AAEJ,QAAQ,CAAC,yBAAyB,EAAE,GAAG,EAAE;IACvC,MAAM,aAAa,GAAG,MAAM,CAAC,KAAK,CAAC;IAEnC,UAAU,CAAC,GAAG,EAAE;QACd,EAAE,CAAC,aAAa,EAAE,CAAC;IACrB,CAAC,CAAC,CAAC;IAEH,SAAS,CAAC,GAAG,EAAE;QACb,MAAM,CAAC,KAAK,GAAG,aAAa,CAAC;IAC/B,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,8DAA8D,EAAE,KAAK,IAAI,EAAE;QAC5E,MAAM,CAAC,GAAG,MAAM,uBAAuB,CAAC,EAAE,CAAC,CAAC;QAC5C,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QAC/B,MAAM,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IACjC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,0CAA0C,EAAE,KAAK,IAAI,EAAE;QACxD,MAAM,EAAE,GAAG,MAAM,uBAAuB,CAAC,EAAE,0BAA0B,EAAE,SAAS,EAAE,CAAC,CAAC;QACpF,MAAM,EAAE,GAAG,MAAM,uBAAuB,CAAC,EAAE,0BAA0B,EAAE,MAAM,EAAE,CAAC,CAAC;QACjF,MAAM,CAAC,EAAE,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QAChC,MAAM,CAAC,EAAE,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IAClC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,8DAA8D,EAAE,KAAK,IAAI,EAAE;QAC5E,MAAM,CAAC,GAAG,MAAM,uBAAuB,CAAC;YACtC,0BAA0B,EAAE,QAAQ;YACpC,sBAAsB,EAAE,SAAS;SAClC,CAAC,CAAC;QACH,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC;QACnC,MAAM,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAClC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,mEAAmE,EAAE,KAAK,IAAI,EAAE;QACjF,MAAM,CAAC,GAAG,MAAM,uBAAuB,CAAC;YACtC,0BAA0B,EAAE,QAAQ;SACrC,CAAC,CAAC;QACH,8DAA8D;QAC9D,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IACjC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,6CAA6C,EAAE,KAAK,IAAI,EAAE;QAC3D,MAAM,CAAC,GAAG,MAAM,uBAAuB,CAAC;YACtC,0BAA0B,EAAE,QAAQ;YACpC,sBAAsB,EAAE,SAAS;YACjC,yBAAyB,EAAE,KAAK;SACjC,CAAC,CAAC;QACH,MAAM,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IACjC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,6DAA6D,EAAE,KAAK,IAAI,EAAE;QAC3E,MAAM,CAAC,KAAK,GAAG,EAAE,CAAC,EAAE,CAAC,KAAK,IAAI,EAAE,CAC9B,IAAI,QAAQ,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,SAAS,EAAE,IAAI,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,MAAM,EAAE,GAAG,EAAE,CAAC,CAC1D,CAAC;QAE7B,MAAM,CAAC,GAAG,MAAM,uBAAuB,CAAC;YACtC,0BAA0B,EAAE,QAAQ;SACrC,CAAC,CAAC;QACH,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC;QACnC,MAAM,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IACjC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,kDAAkD,EAAE,KAAK,IAAI,EAAE;QAChE,MAAM,CAAC,KAAK,GAAG,EAAE,CAAC,EAAE,CAAC,KAAK,IAAI,EAAE;YAC9B,MAAM,IAAI,KAAK,CAAC,cAAc,CAAC,CAAC;QAClC,CAAC,CAA4B,CAAC;QAE9B,MAAM,CAAC,GAAG,MAAM,uBAAuB,CAAC;YACtC,0BAA0B,EAAE,QAAQ;SACrC,CAAC,CAAC;QACH,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IACjC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,kDAAkD,EAAE,KAAK,IAAI,EAAE;QAChE,MAAM,CAAC,GAAG,MAAM,uBAAuB,CAAC;YACtC,0BAA0B,EAAE,UAAU;SACvC,CAAC,CAAC;QACH,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IACjC,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,QAAQ,CAAC,eAAe,EAAE,GAAG,EAAE;IAC7B,EAAE,CAAC,uCAAuC,EAAE,KAAK,IAAI,EAAE;QACrD,MAAM,CAAC,GAAG,MAAM,uBAAuB,CAAC,EAAE,CAAC,CAAC;QAC5C,MAAM,EAAE,GAAG,aAAa,CAAC,CAAC,CAAC,CAAC;QAC5B,MAAM,CAAC,EAAE,CAAC,QAAQ,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QACpC,MAAM,CAAC,EAAE,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAClC,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}

package/dist/src/indexer/indexer-freshness.test.d.ts

@@ -0,0 +1 @@
+export {};

package/dist/src/indexer/indexer-freshness.test.js

@@ -0,0 +1,109 @@
+import { describe, it, expect, beforeEach, afterEach, vi } from "vitest";
+import { mkdtempSync, rmSync, writeFileSync, mkdirSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { Indexer } from "./indexer.js";
+import { getProjectConfig } from "../utils/config.js";
+// Tests for the freshness cache contract (issue #6). The cache exists
+// to keep sverklo_status fast — the disk walk costs ~95ms on a small
+// repo and agents can call status repeatedly in one session. The
+// contract:
+//
+//   1. Result is cached for FRESHNESS_CACHE_MS (2s) after the first
+//      computation.
+//   2. The cache is invalidated by explicit reindex / clearIndex.
+//   3. The file watcher also invalidates on change events (tested via
+//      direct invalidateFreshnessCache() call since we don't want to
+//      stand up chokidar in a unit test).
+describe("Indexer freshness cache", () => {
+    let tmpRoot;
+    let indexer;
+    beforeEach(async () => {
+        tmpRoot = mkdtempSync(join(tmpdir(), "sverklo-freshness-"));
+        // Minimal real repo: one TypeScript file. Indexing this should
+        // take well under a second.
+        mkdirSync(join(tmpRoot, "src"), { recursive: true });
+        writeFileSync(join(tmpRoot, "src", "foo.ts"), "export function foo() { return 42; }\n", "utf-8");
+        const cfg = getProjectConfig(tmpRoot);
+        indexer = new Indexer(cfg);
+        await indexer.index();
+    });
+    afterEach(() => {
+        try {
+            indexer.close();
+        }
+        catch { }
+        try {
+            rmSync(tmpRoot, { recursive: true, force: true });
+        }
+        catch { }
+    });
+    it("returns a freshness result with ageSeconds and dirty/missing lists", () => {
+        const result = indexer.getFreshness();
+        expect(result).toBeDefined();
+        expect(Array.isArray(result.dirtyFiles)).toBe(true);
+        expect(Array.isArray(result.missingFiles)).toBe(true);
+        // Either a number or null — never undefined
+        expect(result.ageSeconds === null || typeof result.ageSeconds === "number").toBe(true);
+    });
+    it("serves from cache on rapid successive calls", () => {
+        // First call: compute. Second call within TTL: serve from cache.
+        // We can't directly observe "did it re-walk?" without instrumentation,
+        // but we can observe that the second call returns the *same*
+        // dirtyFiles array reference if we patch it — or more simply,
+        // measure that 100 calls in a row take negligible time (the cache
+        // hit path is O(1)). A perf threshold works as a regression guard.
+        const t0 = Date.now();
+        for (let i = 0; i < 100; i++) {
+            indexer.getFreshness();
+        }
+        const elapsed = Date.now() - t0;
+        // 100 cached reads should complete in <50ms on any machine that
+        // isn't actively on fire. Real disk walks would take 100× longer.
+        expect(elapsed).toBeLessThan(500);
+    });
+    it("reflects filesystem changes after invalidateFreshnessCache()", () => {
+        const first = indexer.getFreshness();
+        expect(first.dirtyFiles.length).toBe(0);
+        // Add a new file on disk (bypassing the watcher so we control timing)
+        writeFileSync(join(tmpRoot, "src", "bar.ts"), "export const bar = 1;\n", "utf-8");
+        // Without invalidation, the cached result persists for up to 2s.
+        // With invalidation, the next call sees the new file.
+        indexer.invalidateFreshnessCache();
+        const second = indexer.getFreshness();
+        expect(second.dirtyFiles.some((p) => p.includes("bar.ts"))).toBe(true);
+    });
+    it("clearIndex() invalidates the cache", () => {
+        // Establish a cached result first
+        const beforeClear = indexer.getFreshness();
+        expect(beforeClear).toBeDefined();
+        // clearIndex nukes the database and reinitializes. The cache must
+        // be cleared or sverklo_status would report stale dirty/missing
+        // lists against the (now empty) index.
+        indexer.clearIndex();
+        // After clear, the indexed file count is 0 — so the freshness
+        // result should show all on-disk files as dirty (new to the index).
+        const afterClear = indexer.getFreshness();
+        expect(afterClear.dirtyFiles.length).toBeGreaterThan(0);
+        expect(afterClear.missingFiles.length).toBe(0);
+    });
+    it("updates ageSeconds even when serving from cache", () => {
+        // The cache stores the expensive disk-walk result but still
+        // recomputes ageSeconds on every call (wall clock moves on).
+        const first = indexer.getFreshness();
+        const firstAge = first.ageSeconds;
+        expect(typeof firstAge).toBe("number");
+        // Wait a moment and call again. Age should advance even though
+        // the dirty/missing lists are cached.
+        vi.useFakeTimers();
+        vi.advanceTimersByTime(1500);
+        const second = indexer.getFreshness();
+        vi.useRealTimers();
+        // Age on the second call should be higher than the first (by ~1s).
+        // Allow some slack for the test harness timing.
+        if (typeof firstAge === "number" && typeof second.ageSeconds === "number") {
+            expect(second.ageSeconds).toBeGreaterThanOrEqual(firstAge);
+        }
+    });
+});
+//# sourceMappingURL=indexer-freshness.test.js.map

package/dist/src/indexer/indexer-freshness.test.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"indexer-freshness.test.js","sourceRoot":"","sources":["../../../src/indexer/indexer-freshness.test.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,UAAU,EAAE,SAAS,EAAE,EAAE,EAAE,MAAM,QAAQ,CAAC;AACzE,OAAO,EAAE,WAAW,EAAE,MAAM,EAAE,aAAa,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AACxE,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AACjC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,OAAO,EAAE,MAAM,cAAc,CAAC;AACvC,OAAO,EAAE,gBAAgB,EAAE,MAAM,oBAAoB,CAAC;AAEtD,sEAAsE;AACtE,qEAAqE;AACrE,iEAAiE;AACjE,YAAY;AACZ,EAAE;AACF,oEAAoE;AACpE,oBAAoB;AACpB,kEAAkE;AAClE,sEAAsE;AACtE,qEAAqE;AACrE,0CAA0C;AAE1C,QAAQ,CAAC,yBAAyB,EAAE,GAAG,EAAE;IACvC,IAAI,OAAe,CAAC;IACpB,IAAI,OAAgB,CAAC;IAErB,UAAU,CAAC,KAAK,IAAI,EAAE;QACpB,OAAO,GAAG,WAAW,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,oBAAoB,CAAC,CAAC,CAAC;QAE5D,+DAA+D;QAC/D,4BAA4B;QAC5B,SAAS,CAAC,IAAI,CAAC,OAAO,EAAE,KAAK,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACrD,aAAa,CACX,IAAI,CAAC,OAAO,EAAE,KAAK,EAAE,QAAQ,CAAC,EAC9B,wCAAwC,EACxC,OAAO,CACR,CAAC;QAEF,MAAM,GAAG,GAAG,gBAAgB,CAAC,OAAO,CAAC,CAAC;QACtC,OAAO,GAAG,IAAI,OAAO,CAAC,GAAG,CAAC,CAAC;QAC3B,MAAM,OAAO,CAAC,KAAK,EAAE,CAAC;IACxB,CAAC,CAAC,CAAC;IAEH,SAAS,CAAC,GAAG,EAAE;QACb,IAAI,CAAC;YACH,OAAO,CAAC,KAAK,EAAE,CAAC;QAClB,CAAC;QAAC,MAAM,CAAC,CAAA,CAAC;QACV,IAAI,CAAC;YACH,MAAM,CAAC,OAAO,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;QACpD,CAAC;QAAC,MAAM,CAAC,CAAA,CAAC;IACZ,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,oEAAoE,EAAE,GAAG,EAAE;QAC5E,MAAM,MAAM,GAAG,OAAO,CAAC,YAAY,EAAE,CAAC;QACtC,MAAM,CAAC,MAAM,CAAC,CAAC,WAAW,EAAE,CAAC;QAC7B,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACpD,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,YAAY,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACtD,4CAA4C;QAC5C,MAAM,CAAC,MAAM,CAAC,UAAU,KAAK,IAAI,IAAI,OAAO,MAAM,CAAC,UAAU,KAAK,QAAQ,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACzF,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,6CAA6C,EAAE,GAAG,EAAE;QACrD,iEAAiE;QACjE,uEAAuE;QACvE,6DAA6D;QAC7D,8DAA8D;QAC9D,kEAAkE;QAClE,mEAAmE;QACnE,MAAM,EAAE,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QACtB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC;YAC7B,OAAO,CAAC,YAAY,EAAE,CAAC;QACzB,CAAC;QACD,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,EAAE,CAAC;QAChC,gEAAgE;QAChE,kEAAkE;QAClE,MAAM,CAAC,OAAO,CAAC,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC;IACpC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,8DAA8D,EAAE,GAAG,EAAE;QACtE,MAAM,KAAK,GAAG,OAAO,CAAC,YAAY,EAAE,CAAC;QACrC,MAAM,CAAC,KAAK,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAExC,sEAAsE;QACtE,aAAa,CAAC,IAAI,CAAC,OAAO,EAAE,KAAK,EAAE,QAAQ,CAAC,EAAE,yBAAyB,EAAE,OAAO,CAAC,CAAC;QAElF,iEAAiE;QACjE,sDAAsD;QACtD,OAAO,CAAC,wBAAwB,EAAE,CAAC;QACnC,MAAM,MAAM,GAAG,OAAO,CAAC,YAAY,EAAE,CAAC;QACtC,MAAM,CAAC,MAAM,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACzE,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,oCAAoC,EAAE,GAAG,EAAE;QAC5C,kCAAkC;QAClC,MAAM,WAAW,GAAG,OAAO,CAAC,YAAY,EAAE,CAAC;QAC3C,MAAM,CAAC,WAAW,CAAC,CAAC,WAAW,EAAE,CAAC;QAElC,kEAAkE;QAClE,gEAAgE;QAChE,uCAAuC;QACvC,OAAO,CAAC,UAAU,EAAE,CAAC;QAErB,8DAA8D;QAC9D,oEAAoE;QACpE,MAAM,UAAU,GAAG,OAAO,CAAC,YAAY,EAAE,CAAC;QAC1C,MAAM,CAAC,UAAU,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC;QACxD,MAAM,CAAC,UAAU,CAAC,YAAY,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACjD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,iDAAiD,EAAE,GAAG,EAAE;QACzD,4DAA4D;QAC5D,6DAA6D;QAC7D,MAAM,KAAK,GAAG,OAAO,CAAC,YAAY,EAAE,CAAC;QACrC,MAAM,QAAQ,GAAG,KAAK,CAAC,UAAU,CAAC;QAClC,MAAM,CAAC,OAAO,QAAQ,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QAEvC,+DAA+D;QAC/D,sCAAsC;QACtC,EAAE,CAAC,aAAa,EAAE,CAAC;QACnB,EAAE,CAAC,mBAAmB,CAAC,IAAI,CAAC,CAAC;QAC7B,MAAM,MAAM,GAAG,OAAO,CAAC,YAAY,EAAE,CAAC;QACtC,EAAE,CAAC,aAAa,EAAE,CAAC;QAEnB,mEAAmE;QACnE,gDAAgD;QAChD,IAAI,OAAO,QAAQ,KAAK,QAAQ,IAAI,OAAO,MAAM,CAAC,UAAU,KAAK,QAAQ,EAAE,CAAC;YAC1E,MAAM,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,sBAAsB,CAAC,QAAQ,CAAC,CAAC;QAC7D,CAAC;IACH,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}

package/dist/src/indexer/indexer-provider-integration.test.d.ts

@@ -0,0 +1 @@
+export {};

package/dist/src/indexer/indexer-provider-integration.test.js

@@ -0,0 +1,142 @@
+import { describe, it, expect, beforeEach, afterEach, vi } from "vitest";
+import { mkdtempSync, rmSync, writeFileSync, mkdirSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { Indexer } from "./indexer.js";
+import { getProjectConfig } from "../utils/config.js";
+// Integration test that would have caught the issue #9 wiring gap:
+// the provider factory existed and its unit tests passed, but the
+// Indexer never actually called the factory — it imported legacyEmbed
+// directly from ./embedder.js and used it everywhere. Users setting
+// SVERKLO_EMBEDDING_PROVIDER=openai silently got the bundled ONNX
+// model and no visible error.
+//
+// The fix was to lazily initialize the provider on the first index()
+// call and expose it via `indexer.embed()` / `indexer.embeddingProviderName`.
+// These tests lock that wiring in so a refactor can't silently break
+// it again.
+//
+// We don't hit real OpenAI / Ollama endpoints here. The point isn't
+// to prove the providers work (their own unit tests do that). The
+// point is to prove the indexer *uses* whichever provider the env
+// vars selected, instead of always defaulting.
+describe("Indexer + embedding provider integration", () => {
+    let tmpRoot;
+    const originalEnv = { ...process.env };
+    beforeEach(() => {
+        tmpRoot = mkdtempSync(join(tmpdir(), "sverklo-provider-int-"));
+        mkdirSync(join(tmpRoot, "src"), { recursive: true });
+        writeFileSync(join(tmpRoot, "src", "a.ts"), "export function hello() { return 'world'; }\n", "utf-8");
+    });
+    afterEach(() => {
+        try {
+            rmSync(tmpRoot, { recursive: true, force: true });
+        }
+        catch { }
+        // Restore env so tests don't leak into each other.
+        for (const key of Object.keys(process.env)) {
+            if (key.startsWith("SVERKLO_") && !(key in originalEnv)) {
+                delete process.env[key];
+            }
+        }
+        for (const [k, v] of Object.entries(originalEnv)) {
+            if (k.startsWith("SVERKLO_"))
+                process.env[k] = v;
+        }
+    });
+    it("defaults to the bundled provider when no env var is set", async () => {
+        delete process.env.SVERKLO_EMBEDDING_PROVIDER;
+        const indexer = new Indexer(getProjectConfig(tmpRoot));
+        try {
+            await indexer.index();
+            expect(indexer.embeddingProviderName).toBe("default");
+            expect(indexer.embeddingDimensions).toBe(384);
+        }
+        finally {
+            indexer.close();
+        }
+    });
+    it("selects the OpenAI provider when SVERKLO_EMBEDDING_PROVIDER=openai and key is set", async () => {
+        // We can't actually call OpenAI in a test — patch fetch to a
+        // deterministic stub so the provider init succeeds. The indexer
+        // then commits to the openai provider for the rest of its life.
+        process.env.SVERKLO_EMBEDDING_PROVIDER = "openai";
+        process.env.SVERKLO_OPENAI_API_KEY = "sk-test";
+        // Mock fetch so the embedder calls look successful and return
+        // the right number of dimensions per request.
+        const originalFetch = global.fetch;
+        global.fetch = vi.fn(async (_url, init) => {
+            const body = JSON.parse(init.body);
+            const input = Array.isArray(body.input) ? body.input : [body.input];
+            return new Response(JSON.stringify({
+                data: input.map((_, i) => ({
+                    index: i,
+                    embedding: new Array(1536).fill(0),
+                })),
+            }), { status: 200 });
+        });
+        try {
+            const indexer = new Indexer(getProjectConfig(tmpRoot));
+            try {
+                await indexer.index();
+                expect(indexer.embeddingProviderName).toContain("openai");
+                expect(indexer.embeddingDimensions).toBe(1536);
+            }
+            finally {
+                indexer.close();
+            }
+        }
+        finally {
+            global.fetch = originalFetch;
+        }
+    });
+    it("falls back to default when OpenAI is requested but the API key is missing", async () => {
+        process.env.SVERKLO_EMBEDDING_PROVIDER = "openai";
+        delete process.env.SVERKLO_OPENAI_API_KEY;
+        const indexer = new Indexer(getProjectConfig(tmpRoot));
+        try {
+            await indexer.index();
+            // Factory init() throws on missing key → factory falls back to
+            // default. The whole point of the wiring is that this fallback
+            // is visible, not silent.
+            expect(indexer.embeddingProviderName).toBe("default");
+        }
+        finally {
+            indexer.close();
+        }
+    });
+    it("initializes the provider lazily on first embed() if index() hasn't run yet", async () => {
+        delete process.env.SVERKLO_EMBEDDING_PROVIDER;
+        const indexer = new Indexer(getProjectConfig(tmpRoot));
+        try {
+            // Call embed() without first calling index(). The embed method
+            // should fall back to the legacy path and still return vectors
+            // — this covers agents that hit search/recall before the first
+            // index cycle completes.
+            const vecs = await indexer.embed(["hello world"]);
+            expect(Array.isArray(vecs)).toBe(true);
+            expect(vecs.length).toBe(1);
+            expect(vecs[0]).toBeInstanceOf(Float32Array);
+        }
+        finally {
+            indexer.close();
+        }
+    });
+    it("reuses the same provider across multiple index() calls", async () => {
+        delete process.env.SVERKLO_EMBEDDING_PROVIDER;
+        const indexer = new Indexer(getProjectConfig(tmpRoot));
+        try {
+            await indexer.index();
+            const firstName = indexer.embeddingProviderName;
+            // Write a new file and reindex
+            writeFileSync(join(tmpRoot, "src", "b.ts"), "export function goodbye() { return 'world'; }\n", "utf-8");
+            await indexer.index();
+            // Provider identity stays stable — reindex should not reselect.
+            expect(indexer.embeddingProviderName).toBe(firstName);
+        }
+        finally {
+            indexer.close();
+        }
+    });
+});
+//# sourceMappingURL=indexer-provider-integration.test.js.map

package/dist/src/indexer/indexer-provider-integration.test.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"indexer-provider-integration.test.js","sourceRoot":"","sources":["../../../src/indexer/indexer-provider-integration.test.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,UAAU,EAAE,SAAS,EAAE,EAAE,EAAE,MAAM,QAAQ,CAAC;AACzE,OAAO,EAAE,WAAW,EAAE,MAAM,EAAE,aAAa,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AACxE,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AACjC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,OAAO,EAAE,MAAM,cAAc,CAAC;AACvC,OAAO,EAAE,gBAAgB,EAAE,MAAM,oBAAoB,CAAC;AAEtD,mEAAmE;AACnE,kEAAkE;AAClE,sEAAsE;AACtE,oEAAoE;AACpE,kEAAkE;AAClE,8BAA8B;AAC9B,EAAE;AACF,qEAAqE;AACrE,8EAA8E;AAC9E,qEAAqE;AACrE,YAAY;AACZ,EAAE;AACF,oEAAoE;AACpE,kEAAkE;AAClE,kEAAkE;AAClE,+CAA+C;AAE/C,QAAQ,CAAC,0CAA0C,EAAE,GAAG,EAAE;IACxD,IAAI,OAAe,CAAC;IACpB,MAAM,WAAW,GAAG,EAAE,GAAG,OAAO,CAAC,GAAG,EAAE,CAAC;IAEvC,UAAU,CAAC,GAAG,EAAE;QACd,OAAO,GAAG,WAAW,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,uBAAuB,CAAC,CAAC,CAAC;QAC/D,SAAS,CAAC,IAAI,CAAC,OAAO,EAAE,KAAK,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACrD,aAAa,CACX,IAAI,CAAC,OAAO,EAAE,KAAK,EAAE,MAAM,CAAC,EAC5B,+CAA+C,EAC/C,OAAO,CACR,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,SAAS,CAAC,GAAG,EAAE;QACb,IAAI,CAAC;YACH,MAAM,CAAC,OAAO,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;QACpD,CAAC;QAAC,MAAM,CAAC,CAAA,CAAC;QACV,mDAAmD;QACnD,KAAK,MAAM,GAAG,IAAI,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC;YAC3C,IAAI,GAAG,CAAC,UAAU,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,GAAG,IAAI,WAAW,CAAC,EAAE,CAAC;gBACxD,OAAO,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;YAC1B,CAAC;QACH,CAAC;QACD,KAAK,MAAM,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,WAAW,CAAC,EAAE,CAAC;YACjD,IAAI,CAAC,CAAC,UAAU,CAAC,UAAU,CAAC;gBAAE,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;QACnD,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,yDAAyD,EAAE,KAAK,IAAI,EAAE;QACvE,OAAO,OAAO,CAAC,GAAG,CAAC,0BAA0B,CAAC;QAE9C,MAAM,OAAO,GAAG,IAAI,OAAO,CAAC,gBAAgB,CAAC,OAAO,CAAC,CAAC,CAAC;QACvD,IAAI,CAAC;YACH,MAAM,OAAO,CAAC,KAAK,EAAE,CAAC;YACtB,MAAM,CAAC,OAAO,CAAC,qBAAqB,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;YACtD,MAAM,CAAC,OAAO,CAAC,mBAAmB,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAChD,CAAC;gBAAS,CAAC;YACT,OAAO,CAAC,KAAK,EAAE,CAAC;QAClB,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,mFAAmF,EAAE,KAAK,IAAI,EAAE;QACjG,6DAA6D;QAC7D,gEAAgE;QAChE,gEAAgE;QAChE,OAAO,CAAC,GAAG,CAAC,0BAA0B,GAAG,QAAQ,CAAC;QAClD,OAAO,CAAC,GAAG,CAAC,sBAAsB,GAAG,SAAS,CAAC;QAE/C,8DAA8D;QAC9D,8CAA8C;QAC9C,MAAM,aAAa,GAAG,MAAM,CAAC,KAAK,CAAC;QACnC,MAAM,CAAC,KAAK,GAAG,EAAE,CAAC,EAAE,CAAC,KAAK,EAAE,IAAa,EAAE,IAAc,EAAE,EAAE;YAC3D,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAE,IAAyB,CAAC,IAAI,CAAC,CAAC;YACzD,MAAM,KAAK,GAAa,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;YAC9E,OAAO,IAAI,QAAQ,CACjB,IAAI,CAAC,SAAS,CAAC;gBACb,IAAI,EAAE,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC;oBACzB,KAAK,EAAE,CAAC;oBACR,SAAS,EAAE,IAAI,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC;iBACnC,CAAC,CAAC;aACJ,CAAC,EACF,EAAE,MAAM,EAAE,GAAG,EAAE,CAChB,CAAC;QACJ,CAAC,CAA4B,CAAC;QAE9B,IAAI,CAAC;YACH,MAAM,OAAO,GAAG,IAAI,OAAO,CAAC,gBAAgB,CAAC,OAAO,CAAC,CAAC,CAAC;YACvD,IAAI,CAAC;gBACH,MAAM,OAAO,CAAC,KAAK,EAAE,CAAC;gBACtB,MAAM,CAAC,OAAO,CAAC,qBAAqB,CAAC,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC;gBAC1D,MAAM,CAAC,OAAO,CAAC,mBAAmB,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACjD,CAAC;oBAAS,CAAC;gBACT,OAAO,CAAC,KAAK,EAAE,CAAC;YAClB,CAAC;QACH,CAAC;gBAAS,CAAC;YACT,MAAM,CAAC,KAAK,GAAG,aAAa,CAAC;QAC/B,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,2EAA2E,EAAE,KAAK,IAAI,EAAE;QACzF,OAAO,CAAC,GAAG,CAAC,0BAA0B,GAAG,QAAQ,CAAC;QAClD,OAAO,OAAO,CAAC,GAAG,CAAC,sBAAsB,CAAC;QAE1C,MAAM,OAAO,GAAG,IAAI,OAAO,CAAC,gBAAgB,CAAC,OAAO,CAAC,CAAC,CAAC;QACvD,IAAI,CAAC;YACH,MAAM,OAAO,CAAC,KAAK,EAAE,CAAC;YACtB,+DAA+D;YAC/D,+DAA+D;YAC/D,0BAA0B;YAC1B,MAAM,CAAC,OAAO,CAAC,qBAAqB,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QACxD,CAAC;gBAAS,CAAC;YACT,OAAO,CAAC,KAAK,EAAE,CAAC;QAClB,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,4EAA4E,EAAE,KAAK,IAAI,EAAE;QAC1F,OAAO,OAAO,CAAC,GAAG,CAAC,0BAA0B,CAAC;QAE9C,MAAM,OAAO,GAAG,IAAI,OAAO,CAAC,gBAAgB,CAAC,OAAO,CAAC,CAAC,CAAC;QACvD,IAAI,CAAC;YACH,+DAA+D;YAC/D,+DAA+D;YAC/D,+DAA+D;YAC/D,yBAAyB;YACzB,MAAM,IAAI,GAAG,MAAM,OAAO,CAAC,KAAK,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC;YAClD,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACvC,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;YAC5B,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC,YAAY,CAAC,CAAC;QAC/C,CAAC;gBAAS,CAAC;YACT,OAAO,CAAC,KAAK,EAAE,CAAC;QAClB,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,wDAAwD,EAAE,KAAK,IAAI,EAAE;QACtE,OAAO,OAAO,CAAC,GAAG,CAAC,0BAA0B,CAAC;QAE9C,MAAM,OAAO,GAAG,IAAI,OAAO,CAAC,gBAAgB,CAAC,OAAO,CAAC,CAAC,CAAC;QACvD,IAAI,CAAC;YACH,MAAM,OAAO,CAAC,KAAK,EAAE,CAAC;YACtB,MAAM,SAAS,GAAG,OAAO,CAAC,qBAAqB,CAAC;YAEhD,+BAA+B;YAC/B,aAAa,CACX,IAAI,CAAC,OAAO,EAAE,KAAK,EAAE,MAAM,CAAC,EAC5B,iDAAiD,EACjD,OAAO,CACR,CAAC;YACF,MAAM,OAAO,CAAC,KAAK,EAAE,CAAC;YAEtB,gEAAgE;YAChE,MAAM,CAAC,OAAO,CAAC,qBAAqB,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QACxD,CAAC;gBAAS,CAAC;YACT,OAAO,CAAC,KAAK,EAAE,CAAC;QAClB,CAAC;IACH,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}