feat(tui): send /attach images as multimodal content (#2584, #2587) (#2607)

Adds OpenAI-compatible image_url content blocks to the chat message
model, wiring attached images through build_chat_messages_with_reasoning
as multimodal user-content arrays. When images are present, user
messages emit a content array of text + image_url parts instead of a
plain string, matching the OpenAI vision API shape.

- models.rs: new ImageUrlContent struct, ContentBlock::ImageUrl variant
- client/chat.rs: image_parts collection, multimodal wire format for
  user messages, image-aware message inspection, stream-event no-op
- Exhaustiveness arms added across 10 files (compaction, seam_manager,
  capacity_flow, purge, notifications, session_picker, utils,
  working_set, rlm/session, runtime_api)
- Test: request_builder_emits_openai_image_url_parts_for_user_images

Credit: @xyuai (PR #2587 — root cause + initial implementation)
Closes: #2584

Co-authored-by: xyuai <xyuai@users.noreply.github.com>

crates/tui/src/client/chat.rs

@@ -841,6 +841,31 @@ fn message_content_for_inspect(message: &Value) -> String {
     {
         parts.push(content.to_string());
     }
+    if let Some(content) = message.get("content").and_then(Value::as_array) {
+        for part in content {
+            match part.get("type").and_then(Value::as_str) {
+                Some("text") => {
+                    if let Some(text) = part.get("text").and_then(Value::as_str)
+                        && !text.is_empty()
+                    {
+                        parts.push(text.to_string());
+                    }
+                }
+                Some("image_url") => {
+                    let url = part
+                        .get("image_url")
+                        .and_then(|image_url| image_url.get("url"))
+                        .and_then(Value::as_str)
+                        .unwrap_or("");
+                    parts.push(format!(
+                        "[image_url:{}]",
+                        summarize_image_url_for_inspect(url)
+                    ));
+                }
+                _ => {}
+            }
+        }
+    }
     if let Some(reasoning) = message.get("reasoning_content").and_then(Value::as_str)
         && !reasoning.is_empty()
     {
@@ -852,6 +877,13 @@ fn message_content_for_inspect(message: &Value) -> String {
     parts.join("\n")
 }
 
+fn summarize_image_url_for_inspect(url: &str) -> String {
+    let Some((prefix, encoded)) = url.split_once(";base64,") else {
+        return first_chars(url, 96);
+    };
+    format!("{prefix};base64,<{} chars>", encoded.len())
+}
+
 fn tool_result_inspection_for_message(message: &Value) -> Option<ToolResultInspection> {
     if message.get("role").and_then(Value::as_str) != Some("tool") {
         return None;
@@ -1338,6 +1370,7 @@ fn build_chat_messages_with_reasoning(
     for (message_index, message) in messages.iter().enumerate() {
         let role = message.role.as_str();
         let mut text_parts = Vec::new();
+        let mut image_parts = Vec::new();
         let mut thinking_parts = Vec::new();
         let mut tool_calls = Vec::new();
         let mut tool_call_infos = Vec::new();
@@ -1356,6 +1389,14 @@ fn build_chat_messages_with_reasoning(
                         text_parts.push(text.clone());
                     }
                 }
+                ContentBlock::ImageUrl { image_url } => {
+                    image_parts.push(json!({
+                        "type": "image_url",
+                        "image_url": {
+                            "url": image_url.url.clone(),
+                        },
+                    }));
+                }
                 ContentBlock::Thinking { thinking } => thinking_parts.push(thinking.clone()),
                 ContentBlock::ToolUse {
                     id,
@@ -1469,10 +1510,25 @@ fn build_chat_messages_with_reasoning(
             }
         } else if role == "user" {
             let content = text_parts.join("\n");
-            if !content.trim().is_empty() {
+            let has_text = !content.trim().is_empty();
+            let has_images = !image_parts.is_empty();
+            if has_text || has_images {
+                let wire_content = if has_images {
+                    let mut parts = Vec::new();
+                    if has_text {
+                        parts.push(json!({
+                            "type": "text",
+                            "text": content,
+                        }));
+                    }
+                    parts.extend(image_parts);
+                    json!(parts)
+                } else {
+                    json!(content)
+                };
                 let mut msg = json!({
                     "role": "user",
-                    "content": content,
+                    "content": wire_content,
                 });
                 if include_tool_budget_metadata && let Some(turn_meta) = &turn_meta_budget {
                     msg["_turn_meta_budget"] = turn_meta_budget_json(turn_meta);
@@ -2098,6 +2154,7 @@ fn build_stream_events(response: &MessageResponse) -> Vec<StreamEvent> {
                 events.push(StreamEvent::ContentBlockStop { index });
             }
             ContentBlock::ToolResult { .. } => {}
+            ContentBlock::ImageUrl { .. } => {}
             ContentBlock::ServerToolUse { id, name, input } => {
                 events.push(StreamEvent::ContentBlockStart {
                     index,

crates/tui/src/compaction.rs

@@ -265,7 +265,8 @@ fn message_text(msg: &Message) -> String {
             }
             ContentBlock::ServerToolUse { .. }
             | ContentBlock::ToolSearchToolResult { .. }
-            | ContentBlock::CodeExecutionToolResult { .. } => {}
+            | ContentBlock::CodeExecutionToolResult { .. }
+            | ContentBlock::ImageUrl { .. } => {}
         }
     }
     text
@@ -289,7 +290,8 @@ fn extract_paths_from_message(message: &Message, workspace: Option<&Path>) -> Ve
             ContentBlock::Thinking { .. } => Vec::new(),
             ContentBlock::ServerToolUse { .. }
             | ContentBlock::ToolSearchToolResult { .. }
-            | ContentBlock::CodeExecutionToolResult { .. } => Vec::new(),
+            | ContentBlock::CodeExecutionToolResult { .. }
+            | ContentBlock::ImageUrl { .. } => Vec::new(),
         };
         paths.extend(candidates);
     }
@@ -562,7 +564,8 @@ fn estimate_tokens_for_message(message: &Message, include_thinking: bool) -> usi
             ContentBlock::ToolResult { content, .. } => content.len() / 4,
             ContentBlock::ServerToolUse { .. }
             | ContentBlock::ToolSearchToolResult { .. }
-            | ContentBlock::CodeExecutionToolResult { .. } => 0,
+            | ContentBlock::CodeExecutionToolResult { .. }
+            | ContentBlock::ImageUrl { .. } => 0,
         })
         .sum::<usize>()
 }
@@ -1346,7 +1349,8 @@ fn build_formatted_summary_request(
                 }
                 ContentBlock::ServerToolUse { .. }
                 | ContentBlock::ToolSearchToolResult { .. }
-                | ContentBlock::CodeExecutionToolResult { .. } => {}
+                | ContentBlock::CodeExecutionToolResult { .. }
+                | ContentBlock::ImageUrl { .. } => {}
             }
         }
     }

crates/tui/src/core/engine/capacity_flow.rs

@@ -226,7 +226,8 @@ impl Engine {
                     ContentBlock::Thinking { .. }
                     | ContentBlock::ServerToolUse { .. }
                     | ContentBlock::ToolSearchToolResult { .. }
-                    | ContentBlock::CodeExecutionToolResult { .. } => {}
+                    | ContentBlock::CodeExecutionToolResult { .. }
+                    | ContentBlock::ImageUrl { .. } => {}
                 }
             }
         }

crates/tui/src/models.rs

@@ -65,6 +65,12 @@ pub struct SystemBlock {
     pub cache_control: Option<CacheControl>,
 }
 
+/// OpenAI-compatible image URL payload inside a multimodal message.
+#[derive(Debug, Serialize, Deserialize, Clone, PartialEq)]
+pub struct ImageUrlContent {
+    pub url: String,
+}
+
 /// A chat message with role and content blocks.
 #[derive(Debug, Serialize, Deserialize, Clone, PartialEq)]
 pub struct Message {
@@ -82,6 +88,8 @@ pub enum ContentBlock {
         #[serde(skip_serializing_if = "Option::is_none")]
         cache_control: Option<CacheControl>,
     },
+    #[serde(rename = "image_url")]
+    ImageUrl { image_url: ImageUrlContent },
     #[serde(rename = "thinking")]
     Thinking { thinking: String },
     #[serde(rename = "tool_use")]

crates/tui/src/purge.rs

@@ -246,6 +246,7 @@ fn format_content_block(buf: &mut String, blk_idx: usize, block: &ContentBlock)
                 "  [{blk_idx}] CodeExecutionToolResult (id={tool_use_id}, content={snippet})"
             );
         }
+        ContentBlock::ImageUrl { .. } => {}
     }
 }
 

crates/tui/src/rlm/session.rs

@@ -407,6 +407,7 @@ fn compact_content_block(block: &ContentBlock) -> Value {
             "tool_use_id": tool_use_id,
             "content": content,
         }),
+        ContentBlock::ImageUrl { .. } => serde_json::Value::Null,
     }
 }
 

crates/tui/src/runtime_api.rs

@@ -904,6 +904,7 @@ fn session_to_detail(session: SavedSession) -> SessionDetailResponse {
                     crate::models::ContentBlock::CodeExecutionToolResult { tool_use_id, content } => {
                         json!({ "type": "tool_result", "tool_use_id": tool_use_id, "content": content })
                     }
+                    crate::models::ContentBlock::ImageUrl { .. } => serde_json::Value::Null,
                 })
                 .collect();
             json!({

crates/tui/src/seam_manager.rs

@@ -386,7 +386,8 @@ impl SeamManager {
                     }
                     ContentBlock::ServerToolUse { .. }
                     | ContentBlock::ToolSearchToolResult { .. }
-                    | ContentBlock::CodeExecutionToolResult { .. } => {}
+                    | ContentBlock::CodeExecutionToolResult { .. }
+                    | ContentBlock::ImageUrl { .. } => {}
                 }
             }
         }

crates/tui/src/tui/notifications.rs

@@ -699,6 +699,7 @@ pub fn latest_assistant_text(messages: &[Message]) -> Option<String> {
                     | ContentBlock::ServerToolUse { .. }
                     | ContentBlock::ToolSearchToolResult { .. }
                     | ContentBlock::CodeExecutionToolResult { .. } => None,
+                    | ContentBlock::ImageUrl { .. } => None,
                 })
                 .collect::<Vec<_>>()
                 .join("\n");

crates/tui/src/tui/session_picker.rs

@@ -790,6 +790,9 @@ fn message_text_for_history(message: &crate::models::Message) -> String {
             | crate::models::ContentBlock::CodeExecutionToolResult { content, .. } => {
                 format!("tool result: {}", truncate(&content.to_string(), 220))
             }
+            crate::models::ContentBlock::ImageUrl { .. } => {
+                String::from("[image]")
+            }
         };
         let part = part.trim();
         if !part.is_empty() {

crates/tui/src/utils.rs

@@ -499,7 +499,8 @@ pub fn estimate_message_chars(messages: &[Message]) -> usize {
                 ContentBlock::ToolResult { content, .. } => total += content.len(),
                 ContentBlock::ServerToolUse { .. }
                 | ContentBlock::ToolSearchToolResult { .. }
-                | ContentBlock::CodeExecutionToolResult { .. } => {}
+                | ContentBlock::CodeExecutionToolResult { .. }
+                | ContentBlock::ImageUrl { .. } => {}
             }
         }
     }

crates/tui/src/working_set.rs

@@ -1046,7 +1046,8 @@ fn extract_paths_from_message(message: &Message) -> Vec<String> {
             ContentBlock::Thinking { .. }
             | ContentBlock::ServerToolUse { .. }
             | ContentBlock::ToolSearchToolResult { .. }
-            | ContentBlock::CodeExecutionToolResult { .. } => {}
+            | ContentBlock::CodeExecutionToolResult { .. }
+            | ContentBlock::ImageUrl { .. } => {}
         }
     }
     paths
@@ -1211,7 +1212,8 @@ fn message_mentions_any_path(message: &Message, needles: &[String], max_scan_cha
             ContentBlock::Thinking { .. }
             | ContentBlock::ServerToolUse { .. }
             | ContentBlock::ToolSearchToolResult { .. }
-            | ContentBlock::CodeExecutionToolResult { .. } => {}
+            | ContentBlock::CodeExecutionToolResult { .. }
+            | ContentBlock::ImageUrl { .. } => {}
         }
     }
     false