feat(core): configure custom OpenAI embedding endpoints

docs/semantic-search.md

@@ -117,8 +117,8 @@ All settings are fields on `BasicMemoryConfig` and can be set via environment va
 | `milvus_database` | `BASIC_MEMORY_MILVUS_DATABASE` | `"default"` | Milvus database name. |
 | `semantic_embedding_provider` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_PROVIDER` | `"fastembed"` | Embedding provider: `"fastembed"` (local), `"openai"` (API), or `"litellm"` (multi-provider API, **experimental** — advanced users only). |
 | `semantic_embedding_model` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_MODEL` | `"bge-small-en-v1.5"` | Model identifier. FastEmbed models must exist in the installed FastEmbed catalog. Auto-adjusted per provider if left at default. |
-| `semantic_embedding_api_base` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_API_BASE` | Unset | Optional custom endpoint for the LiteLLM provider, including local or self-hosted OpenAI-compatible servers. |
-| `semantic_embedding_api_key` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_API_KEY` | Unset | Optional API key passed directly to the LiteLLM provider. When unset, LiteLLM continues to read provider credential env vars such as `OPENAI_API_KEY`. |
+| `semantic_embedding_api_base` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_API_BASE` | Unset | Optional custom endpoint for the `openai` or `litellm` provider — point either at a local or self-hosted OpenAI-compatible server. |
+| `semantic_embedding_api_key` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_API_KEY` | Unset | Optional API key for the `openai` or `litellm` provider. When unset, `openai` falls back to `OPENAI_API_KEY` and `litellm` reads its provider credential env vars. |
 | `semantic_embedding_dimensions` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_DIMENSIONS` | Provider default | Vector dimensions. Defaults to 384 for FastEmbed and 1536 for OpenAI/LiteLLM OpenAI. Set this to the model's output size when choosing a non-default FastEmbed or LiteLLM model. |
 | `semantic_embedding_forward_dimensions` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_FORWARD_DIMENSIONS` | Auto | LiteLLM-only override for whether configured dimensions are sent as a provider-side output-size request. |
 | `semantic_embedding_batch_size` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_BATCH_SIZE` | `2` | Number of texts to embed per batch. |
@@ -229,6 +229,31 @@ export BASIC_MEMORY_SEMANTIC_EMBEDDING_PROVIDER=openai
 export OPENAI_API_KEY=sk-...
 ```
 
+#### OpenAI-compatible endpoints (local or self-hosted)
+
+The `openai` provider also honors `semantic_embedding_api_base` and
+`semantic_embedding_api_key`, so you can point it at any OpenAI-compatible
+embedding server — llama.cpp (`llama-server --embedding`), vLLM, Text Embeddings
+Inference, LM Studio, or Ollama's OpenAI shim — without moving to the
+experimental LiteLLM provider:
+
+```bash
+export BASIC_MEMORY_SEMANTIC_SEARCH_ENABLED=true
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_PROVIDER=openai
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_MODEL=your-model
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_API_BASE=http://localhost:8080/v1
+# Set the key your server expects; local servers often ignore it, but the
+# OpenAI client still requires a non-empty value.
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_API_KEY=sk-local
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_DIMENSIONS=<model output size>
+bm reindex --embeddings
+```
+
+Set `semantic_embedding_dimensions` to the model's output size — Basic Memory
+creates fixed-dimension vector storage before indexing. The key is optional only
+when the server accepts requests without one and `OPENAI_API_KEY` is set;
+otherwise provide it explicitly.
+
 ### LiteLLM
 
 > **Experimental — advanced users only.** The LiteLLM provider is experimental and aimed at users comfortable operating remote embedding backends: paid API calls, per-model dimension and input-role configuration, and slower reindexing of large corpora. For most users, FastEmbed (local, default) is recommended. See [LiteLLM Provider](litellm-provider.md) for the caveats and tuning.

src/basic_memory/config_models.py

@@ -300,16 +300,17 @@ def __init__(self, **data: Any) -> None: ...
     semantic_embedding_api_base: str | None = Field(
         default=None,
         description=(
-            "Optional custom API base URL for the LiteLLM embedding provider. "
-            "Use this for OpenAI-compatible local or self-hosted embedding servers."
+            "Optional custom API base URL for the openai or litellm embedding "
+            "provider. Use this for OpenAI-compatible local or self-hosted "
+            "embedding servers (llama.cpp, vLLM, TEI, LM Studio, Ollama)."
         ),
     )
     semantic_embedding_api_key: str | None = Field(
         default=None,
         description=(
-            "Optional API key passed directly to the LiteLLM embedding provider. "
-            "When unset, LiteLLM continues to resolve credentials from provider "
-            "environment variables such as OPENAI_API_KEY."
+            "Optional API key passed directly to the openai or litellm embedding "
+            "provider. When unset, the openai provider falls back to OPENAI_API_KEY "
+            "and litellm resolves provider credential environment variables."
         ),
     )
     semantic_embedding_dimensions: int | None = Field(

src/basic_memory/repository/embedding_provider_factory.py

@@ -116,17 +116,17 @@ def _provider_cache_key(app_config: BasicMemoryConfig) -> ProviderCacheKey:
     runtime CPU budget makes the key drift between calls in a container (#872).
     """
     provider_name = app_config.semantic_embedding_provider.strip().lower()
-    litellm_api_base_digest = None
-    litellm_api_key_digest = None
-    if provider_name == "litellm":
-        litellm_api_base_digest = _sensitive_value_digest(app_config.semantic_embedding_api_base)
-        litellm_api_key_digest = _sensitive_value_digest(app_config.semantic_embedding_api_key)
+    api_base_digest = None
+    api_key_digest = None
+    if provider_name in {"openai", "litellm"}:
+        api_base_digest = _sensitive_value_digest(app_config.semantic_embedding_api_base)
+        api_key_digest = _sensitive_value_digest(app_config.semantic_embedding_api_key)
 
     return (
         provider_name,
         app_config.semantic_embedding_model,
-        litellm_api_base_digest,
-        litellm_api_key_digest,
+        api_base_digest,
+        api_key_digest,
         app_config.semantic_embedding_dimensions,
         app_config.semantic_embedding_forward_dimensions,
         app_config.semantic_embedding_batch_size,
@@ -266,6 +266,8 @@ def create_embedding_provider(app_config: BasicMemoryConfig) -> EmbeddingProvide
             model_name = "text-embedding-3-small"
         provider = OpenAIEmbeddingProvider(
             model_name=model_name,
+            api_key=app_config.semantic_embedding_api_key,
+            base_url=app_config.semantic_embedding_api_base,
             batch_size=app_config.semantic_embedding_batch_size,
             request_concurrency=app_config.semantic_embedding_request_concurrency,
             **extra_kwargs,

tests/repository/test_openai_provider.py

@@ -10,6 +10,7 @@
 from basic_memory.config import BasicMemoryConfig
 import basic_memory.repository.embedding_provider_factory as embedding_provider_factory_module
 from basic_memory.repository.embedding_provider_factory import (
+    _provider_cache_key,
     create_embedding_provider,
     reset_embedding_provider_cache,
 )
@@ -179,6 +180,47 @@ def test_embedding_provider_factory_selects_openai_and_applies_default_model():
     assert provider.model_name == "text-embedding-3-small"
 
 
+def test_embedding_provider_factory_forwards_openai_api_configuration():
+    config = BasicMemoryConfig(
+        env="test",
+        projects={"test-project": "/tmp/basic-memory-test"},
+        default_project="test-project",
+        semantic_search_enabled=True,
+        semantic_embedding_provider="openai",
+        semantic_embedding_api_base="https://embedding.example/v1",
+        semantic_embedding_api_key="test-key",
+    )
+
+    provider = create_embedding_provider(config)
+
+    assert isinstance(provider, OpenAIEmbeddingProvider)
+    assert provider._base_url == "https://embedding.example/v1"
+    assert provider._api_key == "test-key"
+
+
+def test_embedding_provider_factory_separates_openai_api_cache_keys():
+    base = dict(
+        env="test",
+        projects={"test-project": "/tmp/basic-memory-test"},
+        default_project="test-project",
+        semantic_search_enabled=True,
+        semantic_embedding_provider="openai",
+        semantic_embedding_api_base="https://one.example/v1",
+        semantic_embedding_api_key="test-key",
+    )
+
+    first = _provider_cache_key(BasicMemoryConfig(**base))
+    second = _provider_cache_key(
+        BasicMemoryConfig(**{**base, "semantic_embedding_api_base": "https://two.example/v1"})
+    )
+    third = _provider_cache_key(
+        BasicMemoryConfig(**{**base, "semantic_embedding_api_key": "other-key"})
+    )
+
+    assert first != second
+    assert first != third
+
+
 def test_embedding_provider_factory_rejects_unknown_provider():
     """Factory should fail fast for unsupported provider names."""
     config = BasicMemoryConfig(