feat: support multimodal image input

.gitignore

@@ -165,6 +165,8 @@ cython_debug/
 
 # Cursor
 .cursor/
+.qoder/
+.claude
 
 # Super Powers
 .superpowers/

README.md

@@ -45,4 +45,4 @@ Reading from stdin is also supported:
 
 ```bash
 echo "Create an OSS Bucket" | iac-code --prompt -
-```
+```

pyproject.toml

@@ -28,6 +28,7 @@ dependencies = [
     "opentelemetry-distro>=0.48b0",
     "opentelemetry-exporter-otlp>=1.27.0",
     "agent-client-protocol>=0.9.0",
+    "pillow==11.0.0",
     "cryptography>=42.0",
     "keyring>=25.0",
     "tree-sitter>=0.23",
@@ -105,6 +106,10 @@ iac-code = "iac_code.cli.main:app"
 [tool.uv]
 index-url = "https://mirrors.aliyun.com/pypi/simple/"
 
+[[tool.uv.index]]
+url = "https://mirrors.aliyun.com/pypi/simple/"
+default = true
+
 [tool.coverage.run]
 source_pkgs = ["iac_code"]
 branch = true

src/iac_code/agent/agent_loop.py

@@ -12,7 +12,7 @@
 
 from loguru import logger
 
-from iac_code.agent.message import TextBlock, ThinkingBlock, ToolResultBlock, ToolUseBlock
+from iac_code.agent.message import ContentBlock, TextBlock, ThinkingBlock, ToolResultBlock, ToolUseBlock
 from iac_code.i18n import _
 from iac_code.services.context_manager import ContextManager
 from iac_code.tools.base import ToolContext, ToolRegistry, ToolResult
@@ -151,20 +151,22 @@ def _get_provider_messages(self):
                                 input=block.get("input"),
                                 content=block.get("content"),
                                 is_error=block.get("is_error", False),
+                                media_type=block.get("media_type"),
+                                data=block.get("data"),
                             )
                         )
                 provider_messages.append(ProviderMessage(role=role, content=blocks))
         return provider_messages
 
-    async def run(self, user_input: str) -> str:
+    async def run(self, user_input: str | list[ContentBlock]) -> str:
         """Non-streaming execution. Returns final text."""
         final_text = ""
         async for event in self.run_streaming(user_input):
             if isinstance(event, TextDeltaEvent):
                 final_text += event.text
         return final_text
 
-    async def run_streaming(self, user_input: str) -> AsyncGenerator[StreamEvent, None]:
+    async def run_streaming(self, user_input: str | list[ContentBlock]) -> AsyncGenerator[StreamEvent, None]:
         """Streaming execution yielding fine-grained StreamEvents.
 
         Flow:
@@ -201,7 +203,15 @@ async def run_streaming(self, user_input: str) -> AsyncGenerator[StreamEvent, No
                 serialize_user_input,
             )
 
-            entry_attrs[GenAiAttr.INPUT_MESSAGES] = serialize_user_input(user_input)
+            # serialize_user_input expects str; for structured input (list[ContentBlock]),
+            # extract text-only segments so telemetry stays readable without leaking image bytes.
+            if isinstance(user_input, str):
+                input_text_for_telemetry = user_input
+            else:
+                input_text_for_telemetry = " ".join(
+                    getattr(b, "text", "") for b in user_input if getattr(b, "type", None) == "text"
+                )
+            entry_attrs[GenAiAttr.INPUT_MESSAGES] = serialize_user_input(input_text_for_telemetry)
             entry_attrs[GenAiAttr.SYSTEM_INSTRUCTIONS] = serialize_system_instructions(self.system_prompt)
 
         with start_span(Spans.ENTRY, entry_attrs) as entry_span:
@@ -248,7 +258,7 @@ async def run_streaming(self, user_input: str) -> AsyncGenerator[StreamEvent, No
                         serialize_output_messages("".join(final_text_chunks), final_stop_reason),
                     )
 
-    async def _run_streaming_inner(self, user_input: str) -> AsyncGenerator[StreamEvent, None]:
+    async def _run_streaming_inner(self, user_input: str | list[ContentBlock]) -> AsyncGenerator[StreamEvent, None]:
         """Inner streaming loop (called from run_streaming inside the ENTRY span)."""
         from iac_code.services.telemetry import start_span
         from iac_code.services.telemetry.names import GenAiAttr, GenAiOperationName, GenAiSpanKind, Spans
@@ -423,7 +433,7 @@ async def _run_streaming_inner(self, user_input: str) -> AsyncGenerator[StreamEv
                         ]
                         self.context_manager.add_tool_results(denied_blocks)
                         if self._session_storage:
-                            from iac_code.agent.message import ContentBlock, Message
+                            from iac_code.agent.message import Message
 
                             denied_content: list[ContentBlock] = list(denied_blocks)
                             self._session_storage.append(
@@ -509,7 +519,7 @@ async def poll_event_queues():
 
                 self.context_manager.add_tool_results(tool_result_blocks)
                 if self._session_storage:
-                    from iac_code.agent.message import ContentBlock, Message
+                    from iac_code.agent.message import Message
 
                     result_content: list[ContentBlock] = list(tool_result_blocks)
                     self._session_storage.append(

src/iac_code/agent/message.py

@@ -44,8 +44,14 @@ class ThinkingBlock(BaseModel):
     thinking: str
 
 
+class ImageBlock(BaseModel):
+    type: Literal["image"] = "image"
+    media_type: str  # 'image/png' | 'image/jpeg' | 'image/gif' | 'image/webp'
+    data: str  # base64
+
+
 # Union type for all content blocks
-ContentBlock = TextBlock | ToolUseBlock | ToolResultBlock | ThinkingBlock
+ContentBlock = TextBlock | ToolUseBlock | ToolResultBlock | ThinkingBlock | ImageBlock
 
 
 class Message(BaseModel):
@@ -110,6 +116,8 @@ def to_api_format(self) -> dict:
                 )
             elif isinstance(block, ThinkingBlock):
                 content_list.append({"type": "thinking", "thinking": block.thinking})
+            elif isinstance(block, ImageBlock):
+                content_list.append({"type": "image", "media_type": block.media_type, "data": block.data})
         return {"role": self.role, "content": content_list}
 
 
@@ -118,7 +126,7 @@ class Conversation(BaseModel):
 
     messages: list[Message] = Field(default_factory=list)
 
-    def add_user_message(self, content: str) -> Message:
+    def add_user_message(self, content: str | list[ContentBlock]) -> Message:
         """Add a user message to the conversation."""
         msg = Message(role="user", content=content)
         self.messages.append(msg)