File size: 9,540 Bytes
87c0663
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0445c27
 
 
 
 
 
87c0663
 
 
 
 
 
0445c27
 
 
 
87c0663
 
 
 
 
 
 
 
 
 
 
 
0445c27
 
 
87c0663
0445c27
 
 
 
 
87c0663
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0445c27
87c0663
0445c27
 
 
87c0663
 
 
0445c27
87c0663
0445c27
 
 
87c0663
 
 
 
 
 
 
 
 
 
 
0445c27
87c0663
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
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
46
47
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
79
80
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
131
132
133
134
135
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
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
import json
import logging
from typing import Any

import litellm

from app.models.domain import AIProviderResponse, ToolCall
from app.tools.registry import ToolRegistry

logger = logging.getLogger(__name__)


class LiteLLMOrchestration:
    """Provides AI orchestration using LiteLLM Router for automatic API management"""

    def __init__(
        self,
        groq_api_key: str,
        groq_model: str,
        openrouter_api_key: str | None = None,
        openrouter_model: str | None = None,
        groq_api_key_2: str | None = None,
        groq_model_2: str | None = None,
        google_ai_api_key: str | None = None,
        google_ai_model: str | None = None,
        max_tool_iterations: int = 3,
        temperature: float = 0.2,
    ) -> None:
        self.name = "litellm-orchestration"
        self.groq_api_key = groq_api_key
        self.groq_model = groq_model
        self.groq_api_key_2 = groq_api_key_2
        self.groq_model_2 = groq_model_2
        self.google_ai_api_key = google_ai_api_key
        self.google_ai_model = google_ai_model
        self.openrouter_api_key = openrouter_api_key
        self.openrouter_model = openrouter_model
        self.max_tool_iterations = max_tool_iterations
        self.temperature = temperature
        self._initialized = False

    @staticmethod
    def _litellm_model(model: str, provider: str) -> str:
        """Ensure model name has the litellm provider prefix.

        litellm strips the provider prefix before sending to the API,
        so 'groq/openai/gpt-oss-120b' sends 'openai/gpt-oss-120b' to Groq.

        Also strips non-litellm prefixes (e.g. 'google/' from OpenRouter
        model IDs) before applying the correct litellm provider prefix.
        """
        for prefix in ("google",):
            if model.startswith(f"{prefix}/"):
                model = model[len(prefix) + 1 :]
                break

        if model.startswith(f"{provider}/"):
            return model
        return f"{provider}/{model}"

    async def _ensure_initialized(self):
        if not self._initialized:
            self.router = litellm.Router(
                model_list=[
                    {
                        "model_name": "groq",
                        "litellm_params": {
                            "model": self._litellm_model(self.groq_model, "groq"),
                            "api_key": self.groq_api_key,
                            "api_base": "https://api.groq.com/openai/v1",
                        },
                    },
                    {
                        "model_name": "groq_2",
                        "litellm_params": {
                            "model": self._litellm_model(self.groq_model_2 or self.groq_model, "groq"),
                            "api_key": self.groq_api_key_2 or self.groq_api_key,
                            "api_base": "https://api.groq.com/openai/v1",
                        },
                    },
                    {
                        "model_name": "google_ai_studio",
                        "litellm_params": {
                            "model": self._litellm_model(self.google_ai_model or self.openrouter_model, "gemini"),
                            "api_key": self.google_ai_api_key or self.openrouter_api_key,
                            "api_base": "https://generativelanguage.googleapis.com/v1beta",
                        },
                    },
                    {
                        "model_name": "openrouter",
                        "litellm_params": {
                            "model": self._litellm_model(self.openrouter_model, "openrouter"),
                            "api_key": self.openrouter_api_key,
                            "api_base": "https://openrouter.ai/api/v1",
                        },
                    },
                ],
                fallbacks=[{"groq": ["groq_2", "google_ai_studio", "openrouter"]},],
                num_retries=0,
                retry_policy={
                    "TimeoutErrorRetries": 0,
                    "RateLimitErrorRetries": 0,
                    "InternalServerErrorRetries": 0,
                },
                routing_strategy="latency-based-routing",
                set_verbose=False,
            )
            self._initialized = True

    async def chat(
        self,
        *,
        messages: list[dict[str, Any]],
        tools: list[dict[str, Any]] | None = None,
        temperature: float | None = None,
        tool_choice: str | None = "auto",
    ) -> AIProviderResponse:
        if temperature is None:
            temperature = self.temperature

        await self._ensure_initialized()

        kwargs = {
            "messages": messages,
            "model": "groq",
            "temperature": temperature,
        }

        if tools:
            kwargs["tools"] = tools
            kwargs["tool_choice"] = tool_choice

        try:
            response = await self.router.acompletion(**kwargs)
            return self._adapt_response(response)
        except Exception as exc:
            logger.error("LiteLLM Router error: %s", exc)
            raise RuntimeError(f"AI request failed: {exc}") from exc

    async def generate_reply(
        self,
        *,
        messages: list[dict[str, Any]],
        tools: list[dict[str, Any]],
        registry: ToolRegistry,
    ) -> str:
        working_messages = [dict(msg) for msg in messages]

        for _ in range(self.max_tool_iterations + 1):
            response = await self.chat(
                messages=working_messages,
                tools=tools,
                temperature=self.temperature,
            )

            if not response.tool_calls:
                content = (response.content or "").strip()
                if content:
                    return content
                continue

            working_messages.append(self._assistant_tool_message(response))
            for tool_call in response.tool_calls:
                result = await self._execute_tool_call(registry, tool_call)
                if result.suppress_llm_reply:
                    return result.error or ""
                working_messages.append(
                    {
                        "role": "tool",
                        "tool_call_id": tool_call.id,
                        "name": tool_call.name,
                        "content": json.dumps(result.to_payload(), ensure_ascii=False),
                    }
                )

        return (
            "I found that this request needs extra checking. "
            "A support team member will follow up with you shortly."
        )

    def _adapt_response(self, response: Any) -> AIProviderResponse:
        raw_message: dict[str, Any] = {}

        if hasattr(response, "choices") and response.choices:
            choice = response.choices[0]
            if hasattr(choice, "message") and choice.message:
                if hasattr(choice.message, "model_dump"):
                    raw_message = choice.message.model_dump(exclude_none=True)
                elif isinstance(choice.message, dict):
                    raw_message = choice.message
                else:
                    raw_message = {
                        "content": getattr(choice.message, "content", None),
                        "tool_calls": getattr(choice.message, "tool_calls", None),
                    }

        tool_calls = []
        for tool_call in raw_message.get("tool_calls") or []:
            if hasattr(tool_call, "function"):
                function = tool_call.function
                name = function.name
                arguments = function.arguments or "{}"
                id_val = getattr(tool_call, "id", None) or f"tool-call-{name}"
            else:
                function = tool_call.get("function") or {}
                name = function.get("name", "")
                arguments = function.get("arguments") or "{}"
                id_val = tool_call.get("id") or f"tool-call-{name}"

            tool_calls.append(
                ToolCall(
                    id=id_val,
                    name=name,
                    arguments=arguments,
                )
            )

        return AIProviderResponse(
            content=raw_message.get("content"),
            tool_calls=tool_calls,
            raw_message=raw_message,
        )

    @staticmethod
    def _assistant_tool_message(response: AIProviderResponse) -> dict[str, Any]:
        if response.raw_message:
            return response.raw_message
        return {
            "role": "assistant",
            "content": response.content,
            "tool_calls": [
                {
                    "id": tool_call.id,
                    "type": "function",
                    "function": {"name": tool_call.name, "arguments": tool_call.arguments},
                }
                for tool_call in response.tool_calls
            ],
        }

    @staticmethod
    async def _execute_tool_call(
        registry: ToolRegistry, tool_call: ToolCall
    ) -> Any:
        try:
            arguments = json.loads(tool_call.arguments or "{}")
            if not isinstance(arguments, dict):
                raise ValueError("Tool arguments must be a JSON object")
        except (json.JSONDecodeError, ValueError) as exc:
            from app.models.domain import ToolResult

            return ToolResult(
                ok=False, data={}, error=f"Invalid tool arguments: {exc}"
            )

        return await registry.execute(tool_call.name, arguments)