Changeset 0.12.0 (#63)

server/providers/provider.go

@@ -52,12 +52,36 @@ type Message struct {
 	Role    string `json:"role"`    // user, assistant, system, tool
 	Content string `json:"content"`
 
+	// Multimodal content parts (v0.12.0+). When set, providers use these
+	// instead of Content for the user message. Content is still set as
+	// a fallback and for message persistence (extracted text summary).
+	ContentParts []ContentPart `json:"content_parts,omitempty"`
+
 	// Tool calling fields (assistant → tool_calls, tool → tool_call_id)
 	ToolCalls  []ToolCall `json:"tool_calls,omitempty"`   // set when assistant requests tools
 	ToolCallID string     `json:"tool_call_id,omitempty"` // set for role="tool" result messages
 	Name       string     `json:"name,omitempty"`         // tool name for role="tool"
 }
 
+// ContentPart is a single element in a multimodal message.
+// Providers convert these to their native wire format.
+type ContentPart struct {
+	Type string `json:"type"` // "text", "image_url", "document"
+
+	// Text content (type="text")
+	Text string `json:"text,omitempty"`
+
+	// Image content (type="image_url") — base64 data URI
+	ImageURL *ImageURL `json:"image_url,omitempty"`
+}
+
+// ImageURL holds image data for multimodal requests.
+// URL is a data URI: "data:image/jpeg;base64,..."
+type ImageURL struct {
+	URL    string `json:"url"`
+	Detail string `json:"detail,omitempty"` // "auto", "low", "high"
+}
+
 // ToolCall represents a function call requested by the LLM.
 type ToolCall struct {
 	ID       string       `json:"id"`