feat: mcp and backfill fixes

frontend/src/lib/api/endpoints.ts

@@ -280,6 +280,10 @@ export function getJob(jobId: number): Promise<JobView> {
   return request<JobView>(`/jobs/${jobId}`, { account: true });
 }
 
+export function cancelJob(jobId: number): Promise<JobView> {
+  return request<JobView>(`/jobs/${jobId}/cancel`, { method: "POST" });
+}
+
 export function listJobs(status?: JobStatus): Promise<JobView[]> {
   return request<JobView[]>("/jobs", {
     account: true,
@@ -350,6 +354,20 @@ export function fetchMedia(
   });
 }
 
+export function transcribeMedia(
+  chatId: number,
+  messageId: number
+): Promise<{ job_id: number }> {
+  return request<{ job_id: number }>("/media/transcribe", {
+    method: "POST",
+    body: {
+      account_id: accounts.selectedId,
+      chat_id: chatId,
+      message_id: messageId,
+    },
+  });
+}
+
 export function listWatches(): Promise<Watch[]> {
   return request<Watch[]>("/watches", { account: true });
 }

frontend/src/lib/api/media.ts

@@ -1,8 +1,11 @@
 import { requestMedia } from "$lib/api/client";
-import { getMessageMedia } from "$lib/api/endpoints";
+import { getMessageMedia, transcribeMedia } from "$lib/api/endpoints";
 import type { MediaRef } from "$lib/api/types";
 import { accounts } from "$lib/stores/accounts.svelte";
 
+const TRANSCRIBE_TRIES = 10;
+const TRANSCRIBE_DELAY = 2000;
+
 export type InlineMedia =
   | {
       state: "ready";
@@ -143,6 +146,43 @@ export function loadMediaItem(media: MediaRef): Promise<InlineMedia> {
   return promise;
 }
 
+function patchTranscript(
+  chatId: number,
+  messageId: number,
+  text: string
+): void {
+  const account = accounts.selectedId;
+  if (account === null) {
+    return;
+  }
+  const cached = ready.get(cacheKey(account, chatId, messageId));
+  if (cached?.state === "ready") {
+    cached.transcript = text;
+  }
+}
+
+export async function requestTranscription(
+  chatId: number,
+  messageId: number
+): Promise<string | null> {
+  await transcribeMedia(chatId, messageId);
+  for (let i = 0; i < TRANSCRIBE_TRIES; i++) {
+    await new Promise((resolve) => {
+      setTimeout(resolve, TRANSCRIBE_DELAY);
+    });
+    try {
+      const meta = await getMessageMedia(chatId, messageId);
+      if (meta.extracted_text) {
+        patchTranscript(chatId, messageId, meta.extracted_text);
+        return meta.extracted_text;
+      }
+    } catch {
+      return null;
+    }
+  }
+  return null;
+}
+
 export function loadInlineMedia(
   chatId: number,
   messageId: number