Use FastEmbed as default embedding - no API key needed

.github/workflows/index.yml

@@ -23,9 +23,11 @@ jobs:
         run: mergelore-index
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-          MERGELORE_EMBED_MODEL: text-embedding-3-small
           MERGELORE_QDRANT_URL: ${{ secrets.QDRANT_URL }}
           QDRANT_API_KEY: ${{ secrets.QDRANT_API_KEY }}
-          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
           GITHUB_REPOSITORY: ${{ github.repository }}
           GITHUB_EVENT_PATH: ${{ github.event_path }}
+          # Default uses local FastEmbed (bge-small-en-v1.5) - no API key needed
+          # To use OpenAI embeddings instead, uncomment:
+          # MERGELORE_EMBED_MODEL: text-embedding-3-small
+          # OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}

indexer/action.yml

@@ -8,8 +8,8 @@ branding:
 
 inputs:
   embed-model:
-    description: "Embedding model: text-embedding-3-small | nomic-embed-text | embed-english-v3.0"
-    default: text-embedding-3-small
+    description: "Embedding model: bge-small-en-v1.5 (default, local) | all-MiniLM-L6-v2 | nomic-embed-text-v1.5 | text-embedding-3-small (OpenAI) | embed-english-v3.0 (Cohere)"
+    default: bge-small-en-v1.5
   qdrant-url:
     description: Qdrant URL (e.g., http://qdrant:6333)
     required: true

indexer/pyproject.toml

@@ -6,14 +6,15 @@ requires-python = ">=3.12"
 license = "MIT"
 dependencies = [
     "qdrant-client>=1.12.0",
-    "openai>=1.60.0",
-    "cohere>=5.0.0",
+    "fastembed>=0.4.0",
     "tiktoken>=0.8.0",
     "PyGithub>=2.5.0",
 ]
 
 [project.optional-dependencies]
-local = ["sentence-transformers>=3.0.0"]
+openai = ["openai>=1.60.0"]
+cohere = ["cohere>=5.0.0"]
+all = ["openai>=1.60.0", "cohere>=5.0.0"]
 dev = ["pytest>=8.0", "pytest-asyncio>=0.24", "pytest-cov", "ruff"]
 
 [project.scripts]

indexer/src/mergelore_indexer/config.py

@@ -5,11 +5,9 @@
 import os
 from dataclasses import dataclass
 
-VALID_EMBED_MODELS = [
-    "text-embedding-3-small",
-    "nomic-embed-text",
-    "embed-english-v3.0",
-]
+from .embed.factory import ALL_MODELS
+
+DEFAULT_EMBED_MODEL = "bge-small-en-v1.5"
 
 
 @dataclass(frozen=True)
@@ -37,11 +35,11 @@ def load_config() -> IndexerConfig:
     if not github_token:
         raise ValueError("GITHUB_TOKEN is required.")
 
-    embed_model = os.environ.get("MERGELORE_EMBED_MODEL", "text-embedding-3-small")
-    if embed_model not in VALID_EMBED_MODELS:
+    embed_model = os.environ.get("MERGELORE_EMBED_MODEL", DEFAULT_EMBED_MODEL)
+    if embed_model not in ALL_MODELS:
         raise ValueError(
             f"Invalid MERGELORE_EMBED_MODEL: '{embed_model}'. "
-            f"Valid options: {', '.join(VALID_EMBED_MODELS)}"
+            f"Valid options: {', '.join(ALL_MODELS)}"
         )
 
     qdrant_url = os.environ.get("MERGELORE_QDRANT_URL", "")

indexer/src/mergelore_indexer/embed/factory.py

@@ -3,32 +3,49 @@
 from __future__ import annotations
 
 from .interface import EmbeddingProvider
-from .openai_embed import OpenAIEmbedding
-from .nomic import NomicEmbedding
-from .cohere_embed import CohereEmbedding
+from .fastembed_embed import FastEmbedEmbedding, MODELS as FASTEMBED_MODELS
+
+FASTEMBED_KEYS = list(FASTEMBED_MODELS.keys())
+
+ALL_MODELS = FASTEMBED_KEYS + ["text-embedding-3-small", "embed-english-v3.0"]
 
 
 def create_embedding_provider(
     model_name: str,
     openai_api_key: str | None = None,
     cohere_api_key: str | None = None,
 ) -> EmbeddingProvider:
+    # FastEmbed models (local, no API key needed)
+    if model_name in FASTEMBED_MODELS:
+        return FastEmbedEmbedding(model_key=model_name)
+
     match model_name:
         case "text-embedding-3-small":
+            try:
+                from .openai_embed import OpenAIEmbedding
+            except ImportError:
+                raise ImportError(
+                    "OpenAI embedding requires the openai package. "
+                    "Install with: pip install 'mergelore-indexer[openai]'"
+                )
             if not openai_api_key:
                 raise ValueError("OPENAI_API_KEY is required for text-embedding-3-small.")
             return OpenAIEmbedding(api_key=openai_api_key)
 
-        case "nomic-embed-text":
-            return NomicEmbedding()
-
         case "embed-english-v3.0":
+            try:
+                from .cohere_embed import CohereEmbedding
+            except ImportError:
+                raise ImportError(
+                    "Cohere embedding requires the cohere package. "
+                    "Install with: pip install 'mergelore-indexer[cohere]'"
+                )
             if not cohere_api_key:
                 raise ValueError("COHERE_API_KEY is required for embed-english-v3.0.")
             return CohereEmbedding(api_key=cohere_api_key)
 
         case _:
             raise ValueError(
                 f"Unknown embedding model: '{model_name}'. "
-                f"Valid options: text-embedding-3-small, nomic-embed-text, embed-english-v3.0"
+                f"Valid options: {', '.join(ALL_MODELS)}"
             )

indexer/src/mergelore_indexer/embed/fastembed_embed.py

@@ -0,0 +1,39 @@
+"""FastEmbed provider - local ONNX-based embedding (no API key needed)."""
+
+from __future__ import annotations
+
+from fastembed import TextEmbedding
+
+
+MODELS = {
+    "bge-small-en-v1.5": {"name": "BAAI/bge-small-en-v1.5", "dimension": 384},
+    "all-MiniLM-L6-v2": {"name": "sentence-transformers/all-MiniLM-L6-v2", "dimension": 384},
+    "nomic-embed-text-v1.5": {"name": "nomic-ai/nomic-embed-text-v1.5", "dimension": 768},
+}
+
+DEFAULT_MODEL = "bge-small-en-v1.5"
+
+
+class FastEmbedEmbedding:
+    def __init__(self, model_key: str = DEFAULT_MODEL) -> None:
+        if model_key not in MODELS:
+            raise ValueError(
+                f"Unknown FastEmbed model: '{model_key}'. "
+                f"Valid options: {', '.join(MODELS.keys())}"
+            )
+        model_info = MODELS[model_key]
+        self._dimension = model_info["dimension"]
+        self._model = TextEmbedding(model_name=model_info["name"])
+
+    @property
+    def name(self) -> str:
+        return "fastembed"
+
+    @property
+    def dimension(self) -> int:
+        return self._dimension
+
+    async def embed(self, texts: list[str]) -> list[list[float]]:
+        # fastembed is sync but fast (ONNX on CPU)
+        embeddings = list(self._model.embed(texts))
+        return [e.tolist() for e in embeddings]

indexer/tests/test_config.py

@@ -17,19 +17,27 @@ def clean_env(monkeypatch):
 
 
 def set_valid_env(monkeypatch):
+    """Set minimal valid env - uses default FastEmbed model, no API keys needed."""
     monkeypatch.setenv("GITHUB_TOKEN", "gh-test")
-    monkeypatch.setenv("MERGELORE_EMBED_MODEL", "text-embedding-3-small")
     monkeypatch.setenv("MERGELORE_QDRANT_URL", "http://qdrant:6333")
-    monkeypatch.setenv("OPENAI_API_KEY", "sk-test")
     monkeypatch.setenv("GITHUB_REPOSITORY", "owner/repo")
 
 
-def test_loads_valid_config(monkeypatch):
+def test_loads_valid_config_with_defaults(monkeypatch):
     set_valid_env(monkeypatch)
     config = load_config()
-    assert config.embed_model == "text-embedding-3-small"
+    assert config.embed_model == "bge-small-en-v1.5"
     assert config.owner == "owner"
     assert config.repo == "repo"
+    assert config.openai_api_key is None
+
+
+def test_loads_openai_model(monkeypatch):
+    set_valid_env(monkeypatch)
+    monkeypatch.setenv("MERGELORE_EMBED_MODEL", "text-embedding-3-small")
+    monkeypatch.setenv("OPENAI_API_KEY", "sk-test")
+    config = load_config()
+    assert config.embed_model == "text-embedding-3-small"
 
 
 def test_missing_github_token(monkeypatch):
@@ -54,26 +62,21 @@ def test_missing_qdrant_url(monkeypatch):
 
 
 def test_openai_model_requires_key(monkeypatch):
-    monkeypatch.setenv("GITHUB_TOKEN", "gh-test")
+    set_valid_env(monkeypatch)
     monkeypatch.setenv("MERGELORE_EMBED_MODEL", "text-embedding-3-small")
-    monkeypatch.setenv("MERGELORE_QDRANT_URL", "http://qdrant:6333")
-    monkeypatch.setenv("GITHUB_REPOSITORY", "owner/repo")
     with pytest.raises(ValueError, match="OPENAI_API_KEY"):
         load_config()
 
 
 def test_cohere_model_requires_key(monkeypatch):
-    monkeypatch.setenv("GITHUB_TOKEN", "gh-test")
+    set_valid_env(monkeypatch)
     monkeypatch.setenv("MERGELORE_EMBED_MODEL", "embed-english-v3.0")
-    monkeypatch.setenv("MERGELORE_QDRANT_URL", "http://qdrant:6333")
-    monkeypatch.setenv("GITHUB_REPOSITORY", "owner/repo")
     with pytest.raises(ValueError, match="COHERE_API_KEY"):
         load_config()
 
 
 def test_missing_github_repository(monkeypatch):
     monkeypatch.setenv("GITHUB_TOKEN", "gh-test")
-    monkeypatch.setenv("MERGELORE_EMBED_MODEL", "nomic-embed-text")
     monkeypatch.setenv("MERGELORE_QDRANT_URL", "http://qdrant:6333")
     with pytest.raises(ValueError, match="GITHUB_REPOSITORY"):
         load_config()

indexer/tests/test_embed_factory.py

@@ -5,28 +5,34 @@
 from mergelore_indexer.embed.factory import create_embedding_provider
 
 
-def test_creates_openai_provider():
-    provider = create_embedding_provider("text-embedding-3-small", openai_api_key="sk-test")
-    assert provider.name == "text-embedding-3-small"
-    assert provider.dimension == 1536
+def test_creates_fastembed_default():
+    provider = create_embedding_provider("bge-small-en-v1.5")
+    assert provider.name == "fastembed"
+    assert provider.dimension == 384
+
+
+def test_creates_fastembed_minilm():
+    provider = create_embedding_provider("all-MiniLM-L6-v2")
+    assert provider.name == "fastembed"
+    assert provider.dimension == 384
+
+
+def test_creates_fastembed_nomic():
+    provider = create_embedding_provider("nomic-embed-text-v1.5")
+    assert provider.name == "fastembed"
+    assert provider.dimension == 768
 
 
 def test_openai_requires_key():
-    with pytest.raises(ValueError, match="OPENAI_API_KEY"):
+    with pytest.raises((ValueError, ImportError)):
         create_embedding_provider("text-embedding-3-small")
 
 
 def test_cohere_requires_key():
-    with pytest.raises(ValueError, match="COHERE_API_KEY"):
+    with pytest.raises((ValueError, ImportError)):
         create_embedding_provider("embed-english-v3.0")
 
 
-def test_creates_cohere_provider():
-    provider = create_embedding_provider("embed-english-v3.0", cohere_api_key="co-test")
-    assert provider.name == "embed-english-v3.0"
-    assert provider.dimension == 1024
-
-
 def test_unknown_model_raises():
     with pytest.raises(ValueError, match="Unknown embedding model"):
         create_embedding_provider("invalid-model")

samples/tier2-qdrant-index.yml

@@ -1,17 +1,20 @@
 # Tier 2: Qdrant vector memory - PR indexer
 # Runs after every merge to main. Extracts the merged PR content,
-# chunks it, embeds it with OpenAI, and stores it in Qdrant.
+# chunks it, embeds it locally with FastEmbed, and stores it in Qdrant.
 # Use this together with tier2-qdrant-review.yml.
 #
+# No embedding API key needed - FastEmbed runs locally on CPU.
+#
 # Secrets needed:
-#   OPENAI_API_KEY  - get one at https://platform.openai.com (for embeddings)
 #   QDRANT_URL      - your Qdrant instance (e.g. https://xyz.cloud.qdrant.io)
 #   QDRANT_API_KEY  - from Qdrant Cloud dashboard
 #
 # Embedding models (set via MERGELORE_EMBED_MODEL):
-#   text-embedding-3-small  - OpenAI, 1536d, ~$0.02/1M tokens (default)
-#   nomic-embed-text        - local CPU, 768d, free
-#   embed-english-v3.0      - Cohere, 1024d, ~$0.01/1M tokens
+#   bge-small-en-v1.5       - FastEmbed, 384d, fast, no API key (default)
+#   all-MiniLM-L6-v2        - FastEmbed, 384d, well-tested, no API key
+#   nomic-embed-text-v1.5   - FastEmbed, 768d, best quality, no API key
+#   text-embedding-3-small  - OpenAI, 1536d, requires OPENAI_API_KEY
+#   embed-english-v3.0      - Cohere, 1024d, requires COHERE_API_KEY
 
 name: mergelore-index
 on:
@@ -30,15 +33,11 @@ jobs:
       - uses: actions/setup-python@v5
         with:
           python-version: "3.12"
-      - name: Install indexer
-        run: pip install mergelore-indexer
-      - name: Index merged PR
-        run: mergelore-index
+      - run: pip install mergelore-indexer
+      - run: mergelore-index
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-          MERGELORE_EMBED_MODEL: text-embedding-3-small
           MERGELORE_QDRANT_URL: ${{ secrets.QDRANT_URL }}
           QDRANT_API_KEY: ${{ secrets.QDRANT_API_KEY }}
-          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
           GITHUB_REPOSITORY: ${{ github.repository }}
           GITHUB_EVENT_PATH: ${{ github.event_path }}