Skip to content

ontocast.tool.vector_store.embedding

Embedding provider abstraction for vector store workflows.

EmbeddingTool

Bases: Tool

Base embedding tool with provider-specific implementations.

Source code in ontocast/tool/vector_store/embedding.py 
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
class EmbeddingTool(Tool):
    """Base embedding tool with provider-specific implementations."""

    config: EmbeddingConfig = Field(default_factory=EmbeddingConfig)

    @abc.abstractmethod
    def embed(self, texts: list[str]) -> list[list[float]]:
        """Return vectors for all given texts."""

    def embed_one(self, text: str) -> list[float]:
        """Return a vector for one text."""
        vectors = self.embed([text])
        if not vectors:
            raise ValueError("Embedding provider returned no vectors for query text")
        return vectors[0]

    @classmethod
    def create(cls, config: EmbeddingConfig) -> "EmbeddingTool":
        """Factory for provider-specific embedding tools."""
        if config.provider == EmbeddingProvider.HUGGINGFACE:
            return HuggingFaceEmbeddingTool(config=config)
        if config.provider == EmbeddingProvider.OPENAI:
            return OpenAIEmbeddingTool(config=config)
        if config.provider == EmbeddingProvider.OLLAMA:
            return OllamaEmbeddingTool(config=config)
        raise ValueError(f"Unsupported embedding provider: {config.provider}")

create(config) classmethod

Factory for provider-specific embedding tools.

Source code in ontocast/tool/vector_store/embedding.py 
36
37
38
39
40
41
42
43
44
45
@classmethod
def create(cls, config: EmbeddingConfig) -> "EmbeddingTool":
    """Factory for provider-specific embedding tools."""
    if config.provider == EmbeddingProvider.HUGGINGFACE:
        return HuggingFaceEmbeddingTool(config=config)
    if config.provider == EmbeddingProvider.OPENAI:
        return OpenAIEmbeddingTool(config=config)
    if config.provider == EmbeddingProvider.OLLAMA:
        return OllamaEmbeddingTool(config=config)
    raise ValueError(f"Unsupported embedding provider: {config.provider}")

embed(texts) abstractmethod

Return vectors for all given texts.

Source code in ontocast/tool/vector_store/embedding.py 
25
26
27
@abc.abstractmethod
def embed(self, texts: list[str]) -> list[list[float]]:
    """Return vectors for all given texts."""

embed_one(text)

Return a vector for one text.

Source code in ontocast/tool/vector_store/embedding.py 
29
30
31
32
33
34
def embed_one(self, text: str) -> list[float]:
    """Return a vector for one text."""
    vectors = self.embed([text])
    if not vectors:
        raise ValueError("Embedding provider returned no vectors for query text")
    return vectors[0]

FastembedBm25SparseTool

Bases: Tool

BM25-style sparse text embeddings via fastembed (Qdrant-compatible).

Source code in ontocast/tool/vector_store/embedding.py 
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
class FastembedBm25SparseTool(Tool):
    """BM25-style sparse text embeddings via fastembed (Qdrant-compatible)."""

    config: EmbeddingConfig = Field(default_factory=EmbeddingConfig)
    _embedder: Any = PrivateAttr(default=None)

    def _get_embedder(self) -> Any:
        if self._embedder is not None:
            return self._embedder
        try:
            fastembed_mod = importlib.import_module("fastembed")
        except ImportError as error:
            raise ImportError(
                "BM25 sparse embeddings require fastembed. "
                "Install it with: uv add 'fastembed[all]'"
            ) from error
        sparse_cls = getattr(fastembed_mod, "SparseTextEmbedding", None)
        if sparse_cls is None:
            raise ImportError("fastembed.SparseTextEmbedding is not available")
        self._embedder = sparse_cls(model_name=self.config.bm25_model_name)
        return self._embedder

    def embed_sparse(self, texts: list[str]) -> list[qdrant_models.SparseVector]:
        """Return Qdrant sparse vectors for all given texts."""
        if not texts:
            return []
        model = self._get_embedder()
        out: list[qdrant_models.SparseVector] = []
        for sparse_emb in model.embed(texts):
            payload = sparse_emb.as_object()
            indices_raw = payload["indices"]
            values_raw = payload["values"]
            indices_list = indices_raw.tolist()
            values_list = values_raw.tolist()
            out.append(
                qdrant_models.SparseVector(
                    indices=[int(i) for i in indices_list],
                    values=[float(v) for v in values_list],
                )
            )
        if len(out) != len(texts):
            raise ValueError("BM25 embedder returned mismatched sparse vector count")
        return out

    def embed_one_sparse(self, text: str) -> qdrant_models.SparseVector:
        vectors = self.embed_sparse([text])
        if not vectors:
            raise ValueError("BM25 embedder returned no sparse vector for query text")
        return vectors[0]

embed_sparse(texts)

Return Qdrant sparse vectors for all given texts.

Source code in ontocast/tool/vector_store/embedding.py 
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
def embed_sparse(self, texts: list[str]) -> list[qdrant_models.SparseVector]:
    """Return Qdrant sparse vectors for all given texts."""
    if not texts:
        return []
    model = self._get_embedder()
    out: list[qdrant_models.SparseVector] = []
    for sparse_emb in model.embed(texts):
        payload = sparse_emb.as_object()
        indices_raw = payload["indices"]
        values_raw = payload["values"]
        indices_list = indices_raw.tolist()
        values_list = values_raw.tolist()
        out.append(
            qdrant_models.SparseVector(
                indices=[int(i) for i in indices_list],
                values=[float(v) for v in values_list],
            )
        )
    if len(out) != len(texts):
        raise ValueError("BM25 embedder returned mismatched sparse vector count")
    return out

HuggingFaceEmbeddingTool

Bases: EmbeddingTool

Local HuggingFace/SentenceTransformer embeddings.

Source code in ontocast/tool/vector_store/embedding.py 
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
class HuggingFaceEmbeddingTool(EmbeddingTool):
    """Local HuggingFace/SentenceTransformer embeddings."""

    _embedder: Any = PrivateAttr(default=None)

    def _get_embedder(self) -> Any:
        if self._embedder is not None:
            return self._embedder
        try:
            sentence_transformers = importlib.import_module("sentence_transformers")
        except ImportError as error:
            raise ImportError(
                "HuggingFace embeddings require sentence-transformers. "
                "Install it with: uv add sentence-transformers"
            ) from error
        self._embedder = sentence_transformers.SentenceTransformer(
            self.config.model_name
        )
        return self._embedder

    def embed(self, texts: list[str]) -> list[list[float]]:
        if not texts:
            return []
        vectors = self._get_embedder().encode(
            texts, convert_to_numpy=True, show_progress_bar=len(texts) > 100
        )
        return [vector.tolist() for vector in vectors]

OllamaEmbeddingTool

Bases: _LangChainEmbeddingTool

Ollama embeddings using either LangChain or direct API fallback.

Source code in ontocast/tool/vector_store/embedding.py 
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
class OllamaEmbeddingTool(_LangChainEmbeddingTool):
    """Ollama embeddings using either LangChain or direct API fallback."""

    def _build_embedder(self) -> Embeddings:
        return OllamaEmbeddings(
            model=self.config.model_name,
            base_url=self.config.base_url,
        )

    def embed(self, texts: list[str]) -> list[list[float]]:
        if not texts:
            return []
        try:
            return super().embed(texts)
        except Exception:
            return self._embed_via_http(texts)

    def _embed_via_http(self, texts: list[str]) -> list[list[float]]:
        base_url = self.config.base_url or "http://localhost:11434"
        endpoint = f"{base_url.rstrip('/')}/api/embeddings"
        vectors: list[list[float]] = []
        with httpx.Client(timeout=30.0) as client:
            for text in texts:
                response = client.post(
                    endpoint,
                    json={"model": self.config.model_name, "prompt": text},
                )
                response.raise_for_status()
                payload = response.json()
                vector = payload.get("embedding")
                if not isinstance(vector, list):
                    raise ValueError(
                        "Ollama embedding response missing 'embedding' vector"
                    )
                vectors.append(vector)
        return vectors

OpenAIEmbeddingTool

Bases: _LangChainEmbeddingTool

OpenAI embeddings via langchain-openai.

Source code in ontocast/tool/vector_store/embedding.py 
 97
 98
 99
100
101
102
103
104
105
106
107
108
class OpenAIEmbeddingTool(_LangChainEmbeddingTool):
    """OpenAI embeddings via langchain-openai."""

    def _build_embedder(self) -> Embeddings:
        api_key = (
            SecretStr(self.config.api_key) if self.config.api_key is not None else None
        )
        return OpenAIEmbeddings(
            model=self.config.model_name,
            openai_api_key=api_key,
            openai_api_base=self.config.base_url,
        )