Skip to content

rag_provider

LlamaIndexEmbeddingWrapper

Bases: BaseEmbedding

Wraps a Wintermute LLMProvider to be used as a LlamaIndex embedding model.

Source code in wintermute/ai/providers/rag_provider.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
75
76
77
78
class LlamaIndexEmbeddingWrapper(BaseEmbedding):
    """Wraps a Wintermute LLMProvider to be used as a LlamaIndex embedding model."""

    _provider: LLMProvider = None  # type: ignore
    _model: str = ""

    def __init__(self, provider: LLMProvider, model_name: str) -> None:
        super().__init__()
        self._provider = provider
        self._model = model_name

    def _get_query_embedding(self, query: str) -> list[float]:
        embeddings = self._provider.embed([query], model=self._model)
        if not embeddings:
            raise ValueError("Embedding provider returned empty list")
        return embeddings[0]

    async def _aget_query_embedding(self, query: str) -> list[float]:
        return self._get_query_embedding(query)

    def _get_text_embedding(self, text: str) -> list[float]:
        embeddings = self._provider.embed([text], model=self._model)
        if not embeddings:
            raise ValueError("Embedding provider returned empty list")
        return embeddings[0]

    async def _aget_text_embedding(self, text: str) -> list[float]:
        return self._get_text_embedding(text)

    def _get_text_embeddings(self, texts: list[str]) -> list[list[float]]:
        return self._provider.embed(texts, model=self._model)

LlamaIndexLLMWrapper

Bases: LLM

Wraps a Wintermute LLMProvider to be used within LlamaIndex.

Source code in wintermute/ai/providers/rag_provider.py 
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
class LlamaIndexLLMWrapper(LLM):
    """Wraps a Wintermute LLMProvider to be used within LlamaIndex."""

    def __init__(self, provider: LLMProvider, model: Optional[str] = None) -> None:
        super().__init__()
        self._provider = provider
        self._model = model

    @property
    def metadata(self) -> Any:
        from llama_index.core.base.llms.types import LLMMetadata

        return LLMMetadata()

    def complete(
        self, prompt: str, formatted: bool = False, **kwargs: Any
    ) -> CompletionResponse:
        req = ChatRequest(
            messages=[Message(role="user", content=prompt)],
            model=self._model,
        )
        resp = self._provider.chat(req)
        return CompletionResponse(text=resp.content)

    def stream_complete(
        self, prompt: str, formatted: bool = False, **kwargs: Any
    ) -> Any:
        raise NotImplementedError("Streaming not supported in wrapper")

    def chat(self, messages: Any, **kwargs: Any) -> Any:
        raise NotImplementedError("Chat not supported in wrapper")

    def stream_chat(self, messages: Any, **kwargs: Any) -> Any:
        raise NotImplementedError("Streaming chat not supported in wrapper")

    async def acomplete(
        self, prompt: str, formatted: bool = False, **kwargs: Any
    ) -> CompletionResponse:
        return self.complete(prompt, formatted, **kwargs)

    async def achat(self, messages: Any, **kwargs: Any) -> Any:
        return self.chat(messages, **kwargs)

    async def astream_chat(self, messages: Any, **kwargs: Any) -> Any:
        return self.stream_chat(messages, **kwargs)

    async def astream_complete(
        self, prompt: str, formatted: bool = False, **kwargs: Any
    ) -> Any:
        return self.stream_complete(prompt, formatted, **kwargs)

RAGProvider

RAG implementation that conforms to LLMProvider.

Source code in wintermute/ai/providers/rag_provider.py 
136
137
138
139
140
141
142
143
144
145
146
147
148
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
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
class RAGProvider:
    """RAG implementation that conforms to LLMProvider."""

    def __init__(
        self,
        name: str,
        base_provider: LLMProvider,
        persist_dir: str,
        embed_provider: Optional[LLMProvider] = None,
        embed_model_id: str = "amazon.titan-embed-text-v2:0",
        description: str = "",
        vector_store: BasePydanticVectorStore | None = None,
    ) -> None:
        self._name = name
        self._description = description or _DEFAULT_RAG_DESCRIPTION
        self.base_provider = base_provider
        self.persist_dir = persist_dir

        # Configure embedding model
        embed_model: Optional[BaseEmbedding] = None
        if embed_provider:
            embed_model = LlamaIndexEmbeddingWrapper(
                provider=embed_provider, model_name=embed_model_id
            )

        # Load index
        if vector_store is not None:
            # External vector store (Qdrant) — index lives in the database
            self.index: VectorStoreIndex = VectorStoreIndex.from_vector_store(
                vector_store,
                embed_model=embed_model,
            )
        else:
            # Local file-based storage (backward compatible)
            storage_context = StorageContext.from_defaults(persist_dir=persist_dir)
            self.index = load_index_from_storage(
                storage_context,
                embed_model=embed_model,
            )  # type: ignore

        # Create local query engine without global Settings
        self.query_engine: BaseQueryEngine = self.index.as_query_engine(
            llm=LlamaIndexLLMWrapper(base_provider),
            embed_model=embed_model,
        )

    @property
    def name(self) -> str:
        return self._name

    @property
    def description(self) -> str:
        return self._description

    def list_models(self) -> list[ModelInfo]:
        return self.base_provider.list_models()

    def chat(self, req: ChatRequest) -> ChatResponse:
        """Augment prompt with RAG and call base provider."""
        if not req.messages:
            return self.base_provider.chat(req)

        last_msg = req.messages[-1].content
        log.info(f"[RAG:{self.name}] Querying local index for: {last_msg[:50]}...")

        # Retrieve context
        response = self.query_engine.query(last_msg)
        context = str(response)

        # Augment the prompt
        augmented_content = (
            f"Context information is below.\n---------------------\n{context}\n---------------------\n"
            f"Using the context above, answer the query.\n"
            f"Query: {last_msg}\nAnswer: "
        )

        # Create new request with augmented message
        new_messages = list(req.messages[:-1])
        new_messages.append(Message(role="user", content=augmented_content))

        new_req = ChatRequest(
            messages=new_messages,
            model=req.model,
            temperature=req.temperature,
            max_tokens=req.max_tokens,
            tools=req.tools,
            tool_choice=req.tool_choice,
            response_format=req.response_format,
            stream=req.stream,
            task_tag=req.task_tag,
        )

        return self.base_provider.chat(new_req)

    def embed(
        self, texts: Iterable[str], model: Optional[str] = None
    ) -> list[list[float]]:
        return self.base_provider.embed(texts, model)

    def count_tokens(self, text: str, model: Optional[str] = None) -> int:
        return self.base_provider.count_tokens(text, model)

chat(req)

Augment prompt with RAG and call base provider.

Source code in wintermute/ai/providers/rag_provider.py 
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
def chat(self, req: ChatRequest) -> ChatResponse:
    """Augment prompt with RAG and call base provider."""
    if not req.messages:
        return self.base_provider.chat(req)

    last_msg = req.messages[-1].content
    log.info(f"[RAG:{self.name}] Querying local index for: {last_msg[:50]}...")

    # Retrieve context
    response = self.query_engine.query(last_msg)
    context = str(response)

    # Augment the prompt
    augmented_content = (
        f"Context information is below.\n---------------------\n{context}\n---------------------\n"
        f"Using the context above, answer the query.\n"
        f"Query: {last_msg}\nAnswer: "
    )

    # Create new request with augmented message
    new_messages = list(req.messages[:-1])
    new_messages.append(Message(role="user", content=augmented_content))

    new_req = ChatRequest(
        messages=new_messages,
        model=req.model,
        temperature=req.temperature,
        max_tokens=req.max_tokens,
        tools=req.tools,
        tool_choice=req.tool_choice,
        response_format=req.response_format,
        stream=req.stream,
        task_tag=req.task_tag,
    )

    return self.base_provider.chat(new_req)