@pentatonic-ai/ai-agent-sdk 0.5.1 → 0.5.2

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pentatonic-ai/ai-agent-sdk",
-  "version": "0.5.1",
+  "version": "0.5.2",
   "description": "TES SDK — LLM observability and lifecycle tracking via Pentatonic Thing Event System. Track token usage, tool calls, and conversations. Manage things through event-sourced lifecycle stages with AI enrichment and vector search.",
   "type": "module",
   "main": "./dist/index.cjs",

package/packages/memory/src/__tests__/api-contract.test.js

@@ -168,6 +168,11 @@ describe("named exports", () => {
 // --- AI client ---
 
 describe("createAIClient", () => {
+  const realFetch = globalThis.fetch;
+  afterEach(() => {
+    globalThis.fetch = realFetch;
+  });
+
   it("returns an object with embed() and chat()", () => {
     const client = createAIClient({
       url: "http://localhost:11434/v1",
@@ -185,6 +190,79 @@ describe("createAIClient", () => {
     });
     expect(client).toBeDefined();
   });
+
+  it("hits /embeddings by default (OpenAI spec)", async () => {
+    let hitUrl;
+    globalThis.fetch = async (url) => {
+      hitUrl = url;
+      return { ok: true, json: async () => ({ data: [{ embedding: [0.1, 0.2] }] }) };
+    };
+    const client = createAIClient({
+      url: "http://localhost:11434/v1",
+      model: "test",
+    });
+    await client.embed("hello");
+    expect(hitUrl).toBe("http://localhost:11434/v1/embeddings");
+  });
+
+  it("uses embeddingPath override (e.g. Pentatonic AI Gateway)", async () => {
+    let hitUrl;
+    globalThis.fetch = async (url) => {
+      hitUrl = url;
+      return { ok: true, json: async () => ({ data: [{ embedding: [0.1] }] }) };
+    };
+    const client = createAIClient({
+      url: "https://lambda-gateway.pentatonic.com/v1",
+      model: "NV-Embed-v2",
+      embeddingPath: "embed",
+    });
+    await client.embed("hello");
+    expect(hitUrl).toBe("https://lambda-gateway.pentatonic.com/v1/embed");
+  });
+
+  it("normalises leading slashes and trailing base-url slashes", async () => {
+    let hitUrl;
+    globalThis.fetch = async (url) => {
+      hitUrl = url;
+      return { ok: true, json: async () => ({ data: [{ embedding: [0.1] }] }) };
+    };
+    const client = createAIClient({
+      url: "https://gateway.test/v1/",
+      model: "m",
+      embeddingPath: "/embed",
+    });
+    await client.embed("hi");
+    expect(hitUrl).toBe("https://gateway.test/v1/embed");
+  });
+
+  it("chatPath override applies to chat() too", async () => {
+    let hitUrl;
+    globalThis.fetch = async (url) => {
+      hitUrl = url;
+      return { ok: true, json: async () => ({ choices: [{ message: { content: "hi" } }] }) };
+    };
+    const client = createAIClient({
+      url: "https://gateway.test/v1",
+      model: "m",
+      chatPath: "chat",
+    });
+    await client.chat([{ role: "user", content: "q" }]);
+    expect(hitUrl).toBe("https://gateway.test/v1/chat");
+  });
+
+  it("chat defaults to /chat/completions", async () => {
+    let hitUrl;
+    globalThis.fetch = async (url) => {
+      hitUrl = url;
+      return { ok: true, json: async () => ({ choices: [{ message: { content: "hi" } }] }) };
+    };
+    const client = createAIClient({
+      url: "http://localhost:11434/v1",
+      model: "m",
+    });
+    await client.chat([{ role: "user", content: "q" }]);
+    expect(hitUrl).toBe("http://localhost:11434/v1/chat/completions");
+  });
 });
 
 // --- Search options contract ---

package/packages/memory/src/ai.js

@@ -8,10 +8,16 @@
 /**
  * Create an AI client from config.
  *
+ * Defaults to OpenAI-standard paths (`/embeddings`, `/chat/completions`).
+ * Override with `embeddingPath` / `chatPath` for gateways that use
+ * different routes — e.g. Pentatonic AI Gateway exposes `/embed`.
+ *
  * @param {object} config
  * @param {string} config.url - Base URL (e.g. "http://ollama:11434/v1")
  * @param {string} config.model - Model name
  * @param {string} [config.apiKey] - Optional API key
+ * @param {string} [config.embeddingPath="embeddings"] - Path appended to url
+ * @param {string} [config.chatPath="chat/completions"] - Path appended to url
  * @param {number} [config.dimensions] - Expected embedding dimensions
  * @returns {object} Client with embed() and chat() methods
  */
@@ -22,6 +28,23 @@ export function createAIClient(config) {
     headers["X-API-Key"] = config.apiKey;
   }
 
+  // Strip leading slashes so callers can use "embed" or "/embed"
+  // interchangeably. Base url may or may not have a trailing slash.
+  // Plain loops (not regex) to avoid polynomial-regex scanner flags.
+  const stripLeading = (s) => {
+    let i = 0;
+    while (i < s.length && s[i] === "/") i++;
+    return i === 0 ? s : s.slice(i);
+  };
+  const stripTrailing = (s) => {
+    let i = s.length;
+    while (i > 0 && s[i - 1] === "/") i--;
+    return i === s.length ? s : s.slice(0, i);
+  };
+  const embeddingPath = stripLeading(config.embeddingPath || "embeddings");
+  const chatPath = stripLeading(config.chatPath || "chat/completions");
+  const baseUrl = stripTrailing(config.url);
+
   return {
     /**
      * Generate an embedding vector for text.
@@ -32,7 +55,7 @@ export function createAIClient(config) {
      */
     async embed(text, inputType = "passage") {
       try {
-        const res = await fetch(`${config.url}/embeddings`, {
+        const res = await fetch(`${baseUrl}/${embeddingPath}`, {
           method: "POST",
           headers,
           body: JSON.stringify({
@@ -70,7 +93,7 @@ export function createAIClient(config) {
      */
     async chat(messages, opts = {}) {
       try {
-        const res = await fetch(`${config.url}/chat/completions`, {
+        const res = await fetch(`${baseUrl}/${chatPath}`, {
           method: "POST",
           headers,
           body: JSON.stringify({

package/packages/memory/src/server.js

@@ -13,6 +13,9 @@
  *   LLM_URL          — OpenAI-compatible chat endpoint (required)
  *   LLM_MODEL        — Chat model name for HyDE (required)
  *   API_KEY          — API key for embedding/LLM endpoints (optional)
+ *   EMBEDDING_PATH   — Path appended to EMBEDDING_URL (default: "embeddings").
+ *                      Set to "embed" for the Pentatonic AI Gateway.
+ *   CHAT_PATH        — Path appended to LLM_URL (default: "chat/completions")
  *   CLIENT_ID        — Client ID for memory scoping (default: "default")
  *   PORT             — HTTP port for SSE transport (default: 3333)
  */
@@ -46,11 +49,13 @@ function createMemory() {
       url: process.env.EMBEDDING_URL,
       model: process.env.EMBEDDING_MODEL,
       apiKey: process.env.API_KEY,
+      embeddingPath: process.env.EMBEDDING_PATH,
     },
     llm: {
       url: process.env.LLM_URL,
       model: process.env.LLM_MODEL,
       apiKey: process.env.API_KEY,
+      chatPath: process.env.CHAT_PATH,
     },
     logger: (msg) => process.stderr.write(`[memory] ${msg}\n`),
   });
@@ -342,7 +347,7 @@ async function main() {
         const health = {
           status: "ok",
           client: CLIENT_ID,
-          version: "0.5.1",
+          version: "0.5.2",
           search: "text",
           db: false,
           ollama: false,