Merge branch 'main' into fix/lmstudio-embeddings-base-url

The embedder fix in this branch left the sibling LLM still hardcoded, so
a self-hosted LM Studio behind a non-default host worked for embeddings
and silently fell back to localhost for generation.

LMStudioConfig baked "http://localhost:1234/v1" into __init__, which
makes any env-var lookup downstream dead code. The field now defaults to
None and LMStudioLLM resolves config value, then LMSTUDIO_BASE_URL, then
the localhost default, matching the embedder and DeepSeek.

mem0-ts/jest.integration.config.js

@@ -2,6 +2,8 @@
 module.exports = {
   ...require("./jest.config"),
   testMatch: ["**/integration/**/*.test.ts"],
+  globalSetup: "<rootDir>/src/client/tests/integration/global-setup.ts",
+  globalTeardown: "<rootDir>/src/client/tests/integration/global-teardown.ts",
   // Run integration tests serially to avoid rate limiting and race conditions
   maxWorkers: 1,
 };

mem0-ts/src/client/tests/integration/global-setup.ts

@@ -0,0 +1,40 @@
+/**
+ * Jest global setup for integration tests.
+ *
+ * Runs a full project cleanup before any integration test starts,
+ * then waits 10 seconds for the async cleanup to propagate.
+ */
+import { MemoryClient } from "../../mem0";
+
+export default async function globalSetup() {
+  const apiKey = process.env.MEM0_API_KEY;
+  if (!apiKey) return; // skip if no key — tests will be skipped too
+
+  const client = new MemoryClient({ apiKey });
+  await client.ping();
+
+  console.log("[integration] Running pre-test cleanup...");
+
+  // Full project wipe — all four filters set explicitly
+  try {
+    await client.deleteAll({
+      userId: "*",
+      agentId: "*",
+      appId: "*",
+      runId: "*",
+    });
+  } catch {
+    // ignore — may 404 if no data exists
+  }
+
+  try {
+    await client.deleteUsers();
+  } catch {
+    // ignore — may throw "No entities to delete"
+  }
+
+  // Wait 10 seconds for async cleanup to propagate
+  console.log("[integration] Waiting 10s for cleanup to propagate...");
+  await new Promise((r) => setTimeout(r, 10_000));
+  console.log("[integration] Pre-test cleanup done.");
+}

mem0-ts/src/client/tests/integration/global-teardown.ts

@@ -0,0 +1,36 @@
+/**
+ * Jest global teardown for integration tests.
+ *
+ * Runs a full project cleanup after all integration tests complete
+ * so no test data is left behind.
+ */
+import { MemoryClient } from "../../mem0";
+
+export default async function globalTeardown() {
+  const apiKey = process.env.MEM0_API_KEY;
+  if (!apiKey) return;
+
+  const client = new MemoryClient({ apiKey });
+  await client.ping();
+
+  console.log("[integration] Running post-test cleanup...");
+
+  try {
+    await client.deleteAll({
+      userId: "*",
+      agentId: "*",
+      appId: "*",
+      runId: "*",
+    });
+  } catch {
+    // ignore
+  }
+
+  try {
+    await client.deleteUsers();
+  } catch {
+    // ignore
+  }
+
+  console.log("[integration] Post-test cleanup done.");
+}

mem0-ts/src/client/tests/integration/helpers.ts

@@ -52,17 +52,13 @@ export async function withRetry<T>(
 }
 
 /**
- * Poll getAll until a user's memories reach minCount and the ID set
+ * Poll getAll until memories appear for a user.
- * stops changing between polls.
+ * The Mem0 API processes memories asynchronously — after add()
- *
+ * we need to wait for them to be available.
- * The Mem0 API processes memories asynchronously, and consolidating a
- * new fact replaces a memory (delete + add) instead of editing it in
- * place. Returning on the first non-empty read hands callers IDs the
- * pipeline then invalidates, so compare IDs rather than length: a
- * replacement leaves the count unchanged.
  *
  * Polls every 15 seconds with a maximum of 4 retries to avoid
- * hitting rate limits. Throws if minCount is never reached.
+ * hitting rate limits. Throws if results aren't available after
+ * all retries.
  */
 export async function waitForMemories(
   client: MemoryClient,
@@ -70,32 +66,18 @@ export async function waitForMemories(
   minCount: number,
   maxRetries = 4,
 ): Promise<Memory[]> {
-  let memories: Memory[] = [];
-  let previousIds = "";
-
   for (let attempt = 1; attempt <= maxRetries; attempt++) {
     const response = await withRetry(() =>
       client.getAll({ filters: { user_id: userId } }),
     );
-    memories = response.results ?? [];
+    const memories = response.results ?? [];
-
+    if (memories.length >= minCount) {
-    const ids = memories
-      .map((m) => m.id)
-      .sort()
-      .join(",");
-    if (memories.length >= minCount && ids === previousIds) {
       return memories;
     }
-    previousIds = ids;
-
     if (attempt < maxRetries) {
       await new Promise((r) => setTimeout(r, 15_000));
     }
   }
-
-  if (memories.length >= minCount) {
-    return memories;
-  }
   throw new Error(
     `waitForMemories: expected at least ${minCount} memories for user "${userId}" but did not get them after ${maxRetries} attempts`,
   );
@@ -220,3 +202,32 @@ export async function cleanupTestUser(
     // ignore
   }
 }
+
+/**
+ * Full project wipe — deletes all memories and all entities.
+ * Equivalent to Python SDK's:
+ *   client.delete_all(user_id="*", agent_id="*", app_id="*", run_id="*")
+ *
+ * Used as cleanup before and after integration test runs so tests
+ * start from a clean slate and don't leave data behind.
+ */
+export async function fullProjectCleanup(client: MemoryClient): Promise<void> {
+  // Delete all memories — all four filters set explicitly
+  try {
+    await client.deleteAll({
+      userId: "*",
+      agentId: "*",
+      appId: "*",
+      runId: "*",
+    });
+  } catch {
+    // ignore — may 404 if no data exists
+  }
+
+  // Delete all entities (users, agents, apps, runs)
+  try {
+    await client.deleteUsers();
+  } catch {
+    // ignore — may throw "No entities to delete"
+  }
+}

mem0-ts/src/client/tests/integration/management.test.ts

@@ -100,7 +100,6 @@ describeIntegration("MemoryClient Integration — Users & Project", () => {
     const hookName = `test-hook-${randomUUID().slice(0, 8)}`;
     const hookUrl = `https://example.com/webhook/${randomUUID().slice(0, 8)}`;
     const updatedName = `updated-hook-${randomUUID().slice(0, 8)}`;
-    const updatedUrl = `https://example.com/updated/${randomUUID().slice(0, 8)}`;
 
     afterAll(async () => {
       if (createdWebhookId) {
@@ -169,7 +168,7 @@ describeIntegration("MemoryClient Integration — Users & Project", () => {
         client.updateWebhook({
           webhookId: createdWebhookId,
           name: updatedName,
-          url: updatedUrl,
+          url: "https://example.com/updated",
           eventTypes: [WebhookEvent.MEMORY_DELETED],
         }),
       );

mem0-ts/src/client/tests/waitForMemories.test.ts

@@ -1,69 +0,0 @@
-import { waitForMemories } from "./integration/helpers";
-
-const clientReturning = (polls: Array<Array<{ id: string }>>) => {
-  let call = 0;
-  return {
-    getAll: jest.fn(async () => ({
-      results: polls[Math.min(call++, polls.length - 1)],
-    })),
-  } as any;
-};
-
-const settle = async <T>(pending: Promise<T>): Promise<T> => {
-  pending.catch(() => {});
-  for (let i = 0; i < 4; i++) {
-    await jest.advanceTimersByTimeAsync(15_000);
-  }
-  return pending;
-};
-
-describe("waitForMemories", () => {
-  beforeEach(() => jest.useFakeTimers());
-  afterEach(() => jest.useRealTimers());
-
-  test("waits for the id set to stop changing", async () => {
-    const client = clientReturning([
-      [{ id: "a" }],
-      [{ id: "a" }, { id: "b" }],
-      [{ id: "a" }, { id: "b" }],
-    ]);
-
-    const memories = await settle(waitForMemories(client, "user", 1));
-
-    expect(memories.map((m) => m.id)).toEqual(["a", "b"]);
-    expect(client.getAll).toHaveBeenCalledTimes(3);
-  });
-
-  test("keeps polling when consolidation replaces a memory at a constant count", async () => {
-    const client = clientReturning([
-      [{ id: "a" }, { id: "b" }],
-      [{ id: "a" }, { id: "c" }],
-      [{ id: "a" }, { id: "c" }],
-    ]);
-
-    const memories = await settle(waitForMemories(client, "user", 1));
-
-    expect(memories.map((m) => m.id)).toEqual(["a", "c"]);
-  });
-
-  test("returns the last read when the set never settles but minCount is met", async () => {
-    const client = clientReturning([
-      [{ id: "a" }],
-      [{ id: "b" }],
-      [{ id: "c" }],
-      [{ id: "d" }],
-    ]);
-
-    const memories = await settle(waitForMemories(client, "user", 1));
-
-    expect(memories.map((m) => m.id)).toEqual(["d"]);
-  });
-
-  test("throws when minCount is never reached", async () => {
-    const client = clientReturning([[]]);
-
-    await expect(settle(waitForMemories(client, "user", 1))).rejects.toThrow(
-      /expected at least 1 memories/,
-    );
-  });
-});

mem0-ts/src/oss/src/embeddings/lmstudio.ts

@@ -12,7 +12,11 @@ export class LMStudioEmbedder implements Embedder {
   private model: string;
 
   constructor(config: EmbeddingConfig) {
-    const baseURL = config.baseURL ?? config.url ?? DEFAULT_BASE_URL;
+    const baseURL =
+      config.baseURL ??
+      config.url ??
+      process.env.LMSTUDIO_BASE_URL ??
+      DEFAULT_BASE_URL;
     const apiKey = config.apiKey || DEFAULT_LMSTUDIO_API_KEY;
     this.openai = new OpenAI({ apiKey, baseURL: String(baseURL) });
     this.model = config.model || DEFAULT_MODEL;

mem0-ts/src/oss/src/llms/lmstudio.ts

@@ -12,7 +12,8 @@ export class LMStudioLLM extends OpenAILLM {
     super({
       ...config,
       apiKey: config.apiKey || DEFAULT_LMSTUDIO_API_KEY,
-      baseURL: config.baseURL ?? DEFAULT_BASE_URL,
+      baseURL:
+        config.baseURL ?? process.env.LMSTUDIO_BASE_URL ?? DEFAULT_BASE_URL,
       model: config.model || DEFAULT_MODEL,
     });
   }

mem0-ts/src/oss/tests/lmstudio-embedder.test.ts

@@ -10,14 +10,64 @@ const mockCreate = jest.fn().mockResolvedValue({
   data: [{ embedding: mockEmbedding }],
 });
 
+let capturedConstructorArgs: any;
+
 jest.mock("openai", () => {
-  return jest.fn().mockImplementation(() => ({
+  return jest.fn().mockImplementation((args: any) => {
-    embeddings: { create: mockCreate },
+    capturedConstructorArgs = args;
-  }));
+    return {
+      embeddings: { create: mockCreate },
+    };
+  });
 });
 
 describe("LMStudioEmbedder (unit)", () => {
-  beforeEach(() => mockCreate.mockClear());
+  beforeEach(() => {
+    mockCreate.mockClear();
+    capturedConstructorArgs = undefined;
+    delete process.env.LMSTUDIO_BASE_URL;
+  });
+
+  afterEach(() => {
+    delete process.env.LMSTUDIO_BASE_URL;
+  });
+
+  it("uses the default base URL when neither config nor env var is set", async () => {
+    const embedder = new LMStudioEmbedder({ model: "test-model" });
+
+    await embedder.embed("hello");
+
+    expect(capturedConstructorArgs).toMatchObject({
+      baseURL: "http://localhost:1234/v1",
+    });
+  });
+
+  it("uses LMSTUDIO_BASE_URL when config does not provide a base URL", async () => {
+    process.env.LMSTUDIO_BASE_URL = "http://example.test:5678/v1";
+
+    const embedder = new LMStudioEmbedder({ model: "test-model" });
+
+    await embedder.embed("hello");
+
+    expect(capturedConstructorArgs).toMatchObject({
+      baseURL: "http://example.test:5678/v1",
+    });
+  });
+
+  it("prefers an explicit config baseURL over LMSTUDIO_BASE_URL", async () => {
+    process.env.LMSTUDIO_BASE_URL = "http://example.test:5678/v1";
+
+    const embedder = new LMStudioEmbedder({
+      model: "test-model",
+      baseURL: "http://config.test:9999/v1",
+    });
+
+    await embedder.embed("hello");
+
+    expect(capturedConstructorArgs).toMatchObject({
+      baseURL: "http://config.test:9999/v1",
+    });
+  });
 
   it("embed() calls OpenAI with encoding_format float and returns vector", async () => {
     const embedder = new LMStudioEmbedder({

mem0-ts/src/oss/tests/lmstudio-llm.test.ts

@@ -5,16 +5,56 @@
 
 import { LMStudioLLM } from "../src/llms/lmstudio";
 
+let capturedConstructorArgs: any;
 const mockCreate = jest.fn();
 
 jest.mock("openai", () => {
-  return jest.fn().mockImplementation(() => ({
+  return jest.fn().mockImplementation((args: any) => {
-    chat: { completions: { create: mockCreate } },
+    capturedConstructorArgs = args;
-  }));
+    return {
+      chat: { completions: { create: mockCreate } },
+    };
+  });
 });
 
 describe("LMStudioLLM (unit)", () => {
-  beforeEach(() => mockCreate.mockClear());
+  beforeEach(() => {
+    mockCreate.mockClear();
+    capturedConstructorArgs = undefined;
+    delete process.env.LMSTUDIO_BASE_URL;
+  });
+
+  afterEach(() => {
+    delete process.env.LMSTUDIO_BASE_URL;
+  });
+
+  it("uses the default base URL when neither config nor env var is set", () => {
+    new LMStudioLLM({});
+
+    expect(capturedConstructorArgs).toMatchObject({
+      baseURL: "http://localhost:1234/v1",
+    });
+  });
+
+  it("uses LMSTUDIO_BASE_URL when config does not provide a base URL", () => {
+    process.env.LMSTUDIO_BASE_URL = "http://example.test:5678/v1";
+
+    new LMStudioLLM({});
+
+    expect(capturedConstructorArgs).toMatchObject({
+      baseURL: "http://example.test:5678/v1",
+    });
+  });
+
+  it("prefers an explicit config baseURL over LMSTUDIO_BASE_URL", () => {
+    process.env.LMSTUDIO_BASE_URL = "http://example.test:5678/v1";
+
+    new LMStudioLLM({ baseURL: "http://config.test:9999/v1" });
+
+    expect(capturedConstructorArgs).toMatchObject({
+      baseURL: "http://config.test:9999/v1",
+    });
+  });
 
   it("generateResponse() returns a text response", async () => {
     mockCreate.mockResolvedValueOnce({

mem0/configs/embeddings/base.py

@@ -34,7 +34,7 @@ class BaseEmbedderConfig(ABC):
         # Gemini specific
         output_dimensionality: Optional[str] = None,
         # LM Studio specific
-        lmstudio_base_url: Optional[str] = "http://localhost:1234/v1",
+        lmstudio_base_url: Optional[str] = None,
         # AWS Bedrock specific
         aws_access_key_id: Optional[str] = None,
         aws_secret_access_key: Optional[str] = None,
@@ -70,7 +70,8 @@ class BaseEmbedderConfig(ABC):
         :type memory_update_embedding_type: Optional[str], optional
         :param memory_search_embedding_type: The type of embedding to use for the search memory action, defaults to None
         :type memory_search_embedding_type: Optional[str], optional
-        :param lmstudio_base_url: LM Studio base URL to be use, defaults to "http://localhost:1234/v1"
+        :param lmstudio_base_url: LM Studio base URL to be use, defaults to None
+            (resolved by the provider from LMSTUDIO_BASE_URL, then "http://localhost:1234/v1")
         :type lmstudio_base_url: Optional[str], optional
         """
 

mem0/configs/llms/lmstudio.py

@@ -39,6 +39,7 @@ class LMStudioConfig(BaseLlmConfig):
             vision_details: Vision detail level, defaults to "auto"
             http_client_proxies: HTTP client proxy settings, defaults to None
             lmstudio_base_url: LM Studio base URL, defaults to None
+                (resolved by the provider from LMSTUDIO_BASE_URL, then "http://localhost:1234/v1")
             lmstudio_response_format: LM Studio response format, defaults to None
         """
         # Initialize base parameters
@@ -55,5 +56,5 @@ class LMStudioConfig(BaseLlmConfig):
         )
 
         # LM Studio-specific parameters
-        self.lmstudio_base_url = lmstudio_base_url or "http://localhost:1234/v1"
+        self.lmstudio_base_url = lmstudio_base_url
         self.lmstudio_response_format = lmstudio_response_format

mem0/embeddings/lmstudio.py

@@ -1,3 +1,4 @@
+import os
 from typing import Literal, Optional
 
 from openai import OpenAI
@@ -13,6 +14,9 @@ class LMStudioEmbedding(EmbeddingBase):
         self.config.model = self.config.model or "nomic-ai/nomic-embed-text-v1.5-GGUF/nomic-embed-text-v1.5.f16.gguf"
         self.config.embedding_dims = self.config.embedding_dims or 1536
         self.config.api_key = self.config.api_key or "lm-studio"
+        self.config.lmstudio_base_url = (
+            self.config.lmstudio_base_url or os.getenv("LMSTUDIO_BASE_URL") or "http://localhost:1234/v1"
+        )
 
         self.client = OpenAI(base_url=self.config.lmstudio_base_url, api_key=self.config.api_key)
 

mem0/llms/lmstudio.py

@@ -1,4 +1,5 @@
 import json
+import os
 from typing import Dict, List, Optional, Union
 
 from openai import OpenAI
@@ -37,6 +38,9 @@ class LMStudioLLM(LLMBase):
             or "lmstudio-community/Meta-Llama-3.1-70B-Instruct-GGUF/Meta-Llama-3.1-70B-Instruct-IQ2_M.gguf"
         )
         self.config.api_key = self.config.api_key or "lm-studio"
+        self.config.lmstudio_base_url = (
+            self.config.lmstudio_base_url or os.getenv("LMSTUDIO_BASE_URL") or "http://localhost:1234/v1"
+        )
 
         self.client = OpenAI(base_url=self.config.lmstudio_base_url, api_key=self.config.api_key)
 

tests/embeddings/test_lm_studio_embeddings.py

@@ -6,7 +6,6 @@ from mem0.configs.embeddings.base import BaseEmbedderConfig
 from mem0.embeddings.lmstudio import LMStudioEmbedding
 
 
-
 @pytest.fixture
 def mock_lm_studio_client():
     with patch("mem0.embeddings.lmstudio.OpenAI") as mock_openai:
@@ -80,3 +79,31 @@ def test_embed_batch_count_mismatch_raises(mock_lm_studio_client):
 
     with pytest.raises(ValueError, match="returned 1 embeddings for 2 texts"):
         embedder.embed_batch(["first text", "second text"])
+
+
+def test_base_url_honors_lmstudio_base_url_env(mock_lm_studio_client, monkeypatch):
+    monkeypatch.setenv("LMSTUDIO_BASE_URL", "http://lmstudio.internal:9000/v1")
+
+    with patch("mem0.embeddings.lmstudio.OpenAI") as mock_openai:
+        LMStudioEmbedding(BaseEmbedderConfig())
+
+    assert mock_openai.call_args.kwargs["base_url"] == "http://lmstudio.internal:9000/v1"
+
+
+def test_base_url_prefers_explicit_config_over_env(mock_lm_studio_client, monkeypatch):
+    monkeypatch.setenv("LMSTUDIO_BASE_URL", "http://lmstudio.internal:9000/v1")
+
+    with patch("mem0.embeddings.lmstudio.OpenAI") as mock_openai:
+        LMStudioEmbedding(BaseEmbedderConfig(lmstudio_base_url="http://explicit:1234/v1"))
+
+    assert mock_openai.call_args.kwargs["base_url"] == "http://explicit:1234/v1"
+
+
+def test_base_url_falls_back_to_localhost_default(mock_lm_studio_client, monkeypatch):
+    monkeypatch.delenv("LMSTUDIO_BASE_URL", raising=False)
+
+    with patch("mem0.embeddings.lmstudio.OpenAI") as mock_openai:
+        embedder = LMStudioEmbedding(BaseEmbedderConfig())
+
+    assert mock_openai.call_args.kwargs["base_url"] == "http://localhost:1234/v1"
+    assert embedder.config.lmstudio_base_url == "http://localhost:1234/v1"

tests/llms/test_lm_studio.py

@@ -2,6 +2,7 @@ from unittest.mock import Mock, patch
 
 import pytest
 
+from mem0.configs.llms.base import BaseLlmConfig
 from mem0.configs.llms.lmstudio import LMStudioConfig
 from mem0.llms.lmstudio import LMStudioLLM
 
@@ -70,3 +71,40 @@ def test_generate_response_specifying_response_format(mock_lm_studio_client):
     )
 
     assert response == "I'm doing well, thank you for asking!"
+
+
+def test_base_url_honors_lmstudio_base_url_env(monkeypatch):
+    monkeypatch.setenv("LMSTUDIO_BASE_URL", "http://lmstudio.internal:9000/v1")
+
+    with patch("mem0.llms.lmstudio.OpenAI") as mock_openai:
+        LMStudioLLM(LMStudioConfig())
+
+    assert mock_openai.call_args.kwargs["base_url"] == "http://lmstudio.internal:9000/v1"
+
+
+def test_base_url_prefers_explicit_config_over_env(monkeypatch):
+    monkeypatch.setenv("LMSTUDIO_BASE_URL", "http://lmstudio.internal:9000/v1")
+
+    with patch("mem0.llms.lmstudio.OpenAI") as mock_openai:
+        LMStudioLLM(LMStudioConfig(lmstudio_base_url="http://explicit:1234/v1"))
+
+    assert mock_openai.call_args.kwargs["base_url"] == "http://explicit:1234/v1"
+
+
+def test_base_url_falls_back_to_localhost_default(monkeypatch):
+    monkeypatch.delenv("LMSTUDIO_BASE_URL", raising=False)
+
+    with patch("mem0.llms.lmstudio.OpenAI") as mock_openai:
+        llm = LMStudioLLM(LMStudioConfig())
+
+    assert mock_openai.call_args.kwargs["base_url"] == "http://localhost:1234/v1"
+    assert llm.config.lmstudio_base_url == "http://localhost:1234/v1"
+
+
+def test_base_url_honors_env_via_base_llm_config(monkeypatch):
+    monkeypatch.setenv("LMSTUDIO_BASE_URL", "http://lmstudio.internal:9000/v1")
+
+    with patch("mem0.llms.lmstudio.OpenAI") as mock_openai:
+        LMStudioLLM(BaseLlmConfig(model="some-model"))
+
+    assert mock_openai.call_args.kwargs["base_url"] == "http://lmstudio.internal:9000/v1"