feat: add cluster-agent endpoint, fix OCR labeling, fix ingestion blocking

- Add POST /api/v1/file/:file_uuid/cluster-agent endpoint for on-demand face clustering
- Fix OCR chunks being labeled as ASRX: use ChunkType.as_str() instead of {:?}
- Rule 1 now deletes old chunks before re-inserting to avoid stale data
- Add fallback face_traced.json when store_traced_faces.py fails
- ingestion_complete now handles status='error' to unblock jobs
- VLM describe tool now uses trace_id with pre-extracted face crops

src/core/agent/tools.rs

@@ -1,5 +1,6 @@
 use base64::{engine::general_purpose::STANDARD as BASE64, Engine};
 use serde_json;
+use std::time::Duration;
 
 use crate::core::db::qdrant_db::QdrantDb;
 use crate::core::db::schema;
@@ -1233,3 +1234,115 @@ pub async fn exec_search_by_appearance(
         Err("Color search output not found".to_string())
     }
 }
+
+fn face_crop_path(file_uuid: &str, trace_id: i32) -> Option<std::path::PathBuf> {
+    let base = std::env::var("MOMENTRY_OUTPUT_DIR")
+        .unwrap_or_else(|_| "/Users/accusys/momentry/output".to_string());
+    let dir = std::path::PathBuf::from(base)
+        .join(".faces")
+        .join(file_uuid)
+        .join(trace_id.to_string());
+    if !dir.exists() {
+        return None;
+    }
+    let mut entries: Vec<_> = match std::fs::read_dir(&dir) {
+        Ok(e) => e.filter_map(|e| e.ok()).collect(),
+        Err(_) => return None,
+    };
+    entries.sort_by_key(|e| e.file_name());
+    entries.first().map(|e| e.path())
+}
+
+pub async fn exec_vlm_describe(
+    pool: &sqlx::PgPool,
+    args: &serde_json::Value,
+) -> Result<String, String> {
+    let file_uuid = args.get("file_uuid").and_then(|v| v.as_str()).unwrap_or("");
+    let trace_id = args.get("trace_id").and_then(|v| v.as_i64()).unwrap_or(0) as i32;
+    let prompt = args
+        .get("prompt")
+        .and_then(|v| v.as_str())
+        .unwrap_or("Describe this person's clothing and appearance. Focus on colors, clothing type, and any distinctive visual features.");
+
+    if file_uuid.is_empty() {
+        return Ok(serde_json::json!({"error": "file_uuid is required"}).to_string());
+    }
+    if trace_id <= 0 {
+        return Ok(serde_json::json!({"error": "trace_id is required and must be > 0"}).to_string());
+    }
+
+    let crop_path = face_crop_path(file_uuid, trace_id)
+        .ok_or_else(|| format!("No face crop found for {} trace {}", file_uuid, trace_id))?;
+
+    let jpeg_bytes = std::fs::read(&crop_path)
+        .map_err(|e| format!("Failed to read face crop: {}", e))?;
+
+    let videos = schema::table_name("videos");
+    let fps: f64 = sqlx::query_scalar(&format!(
+        "SELECT COALESCE(fps, 25.0) FROM {} WHERE file_uuid = $1",
+        videos
+    ))
+    .bind(file_uuid)
+    .fetch_optional(pool)
+    .await
+    .map_err(|e| e.to_string())?
+    .unwrap_or(25.0);
+
+    let frame_name = crop_path
+        .file_stem()
+        .and_then(|s| s.to_str())
+        .and_then(|s| s.parse::<i64>().ok())
+        .unwrap_or(0);
+    let timestamp_secs = frame_name as f64 / fps;
+
+    let base64_img = BASE64.encode(&jpeg_bytes);
+
+    let ollama_url = std::env::var("OLLAMA_URL")
+        .unwrap_or_else(|_| "http://localhost:11434".to_string());
+    let model = std::env::var("VLM_MODEL").unwrap_or_else(|_| "llava".to_string());
+
+    let body = serde_json::json!({
+        "model": model,
+        "prompt": prompt,
+        "images": [base64_img],
+        "stream": false,
+        "options": {
+            "num_predict": 80
+        }
+    });
+
+    let client = reqwest::Client::builder()
+        .timeout(Duration::from_secs(30))
+        .build()
+        .map_err(|e| format!("Failed to create HTTP client: {}", e))?;
+
+    let resp = client
+        .post(format!("{}/api/generate", ollama_url))
+        .json(&body)
+        .send()
+        .await
+        .map_err(|e| format!("Ollama request failed: {}", e))?;
+
+    let resp_json: serde_json::Value = resp
+        .json()
+        .await
+        .map_err(|e| format!("Failed to parse Ollama response: {}", e))?;
+
+    let description = resp_json
+        .get("response")
+        .and_then(|v| v.as_str())
+        .unwrap_or("No description returned")
+        .to_string();
+
+    Ok(serde_json::json!({
+        "tool": "vlm_describe",
+        "result": {
+            "file_uuid": file_uuid,
+            "trace_id": trace_id,
+            "frame": frame_name,
+            "description": description,
+            "time_sec": (timestamp_secs * 100.0).round() / 100.0
+        }
+    })
+    .to_string())
+}