Added reference resolver WIP (codex)

agent.py

@@ -1,9 +1,16 @@
+import base64
+import mimetypes
 from dataclasses import dataclass
-from typing import Literal
+from pathlib import Path
+from typing import Literal, cast
 from pydantic import BaseModel
 
 import httpx2
 from openai import OpenAI
+from openai.types.chat import (
+    ChatCompletionContentPartParam,
+    ChatCompletionMessageParam,
+)
 
 
 @dataclass
@@ -28,17 +35,70 @@ class Agent:
 
     def completion(self, messages: list[AgentMessage], **kwargs) -> str:
         """Generate a completion for specified messages."""
-        messages_raw = []
+        messages_raw: list[ChatCompletionMessageParam] = []
         for m in messages:
             messages_raw.append(
-                {
+                cast(ChatCompletionMessageParam, {
                     "role": m.role,
                     "content": m.content
-                }
+                })
             )
         response = self._client.chat.completions.parse(
             model=self._model,
             messages=messages_raw,
             **kwargs
         )
-        return response.choices[0].message.content # type: ignore
+        return response.choices[0].message.content # type: ignore
+
+    def completion_with_images(
+        self,
+        messages: list[AgentMessage],
+        image_paths: list[str],
+        *,
+        detail: Literal["low", "high", "auto"] = "auto",
+        **kwargs,
+    ) -> str:
+        """Generate a completion with local images attached to the last message."""
+        if not messages or messages[-1].role != "user":
+            raise ValueError("The last message must be a user message")
+        if not image_paths:
+            raise ValueError("At least one image is required")
+
+        messages_raw: list[ChatCompletionMessageParam] = [
+            cast(ChatCompletionMessageParam, {
+                "role": message.role,
+                "content": message.content,
+            })
+            for message in messages[:-1]
+        ]
+        content: list[ChatCompletionContentPartParam] = [
+            {"type": "text", "text": messages[-1].content}
+        ]
+        for index, image_path in enumerate(image_paths, start=1):
+            path = Path(image_path)
+            mime_type = mimetypes.guess_type(path.name)[0] or "image/jpeg"
+            if not mime_type.startswith("image/"):
+                raise ValueError(f"Unsupported image type: {image_path}")
+            encoded = base64.b64encode(path.read_bytes()).decode("ascii")
+            content.append({
+                "type": "text",
+                "text": f"Изображение {index}",
+            })
+            content.append({
+                "type": "image_url",
+                "image_url": {
+                    "url": f"data:{mime_type};base64,{encoded}",
+                    "detail": detail,
+                },
+            })
+        messages_raw.append({
+            "role": "user",
+            "content": content,
+        })
+
+        response = self._client.chat.completions.parse(
+            model=self._model,
+            messages=messages_raw,
+            **kwargs,
+        )
+        return response.choices[0].message.content # type: ignore