attach a PDF and the agent reads it; the microphone button is gone

Two requests, one file apart.

── The microphone ──

Removed from the chat composer. The HOOK stays: `useAudioRecording` and `blobToWav`
are also the file browser's DictateDialog, which is not what was asked to go.

── PDFs, by not extracting anything ──

A PDF entry sat in the attach menu once with no `onSelect` at all — the menu offered
it and did nothing, because nothing here could extract PDF text. It is back and real,
and it still extracts nothing: the file is uploaded, left on disk, and the agent is
given its PATH. Its own `Read` handles PDFs by page, better than any extraction this
could have shipped, and it can go back for a section later without paying for it
twice.

The alternative was base64 in the message. That is megabytes in the turn, re-sent for
the rest of the conversation, to produce worse text.

── What this uncovered ──

`attachmentIds` has been on this websocket and in `types.ts` for a long time and
NOTHING on the server read it. Images work because they travel separately as base64,
so the ids were dead on the wire — and a PDF is exactly the case that needs them. They
resolve to paths now, appended to the prompt as `Attached files (read from disk)`.

Ids arrive in a websocket message and become a filesystem path, so they are checked
rather than trusted: only the `<uuid>.<ext>` shape uploads actually generate, and only
in the two directories uploads actually write to. Verified — a valid id resolves, an
unknown one yields nothing, and `../../../../etc/passwd` and `; rm -rf /` yield
nothing.

Both upload directories are searched because the first message of a chat carries ids
written to `tmp/` (no session yet) and later ones carry ids under the session.

PDFs may be 25 MB against an image's 5: nothing carries them into a prompt, so the
ceiling is storage rather than context.

tsgo clean, frontend builds, 914 pass / 7 fail unchanged.

`[open]` For a MEMBER the file lands under `data/<email>/attachments/`, and whether
their Linux user can read it depends on the confinement ACLs. Untested — the owner
path is what was verified.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

src/servers/api/chat/attached-files.ts

@@ -0,0 +1,48 @@
+import { existsSync } from 'node:fs';
+import { join } from 'node:path';
+import { getAttachmentsDir, getTmpAttachmentsDir } from '../../data-path';
+import { getUserById } from 'officerdb';
+
+/**
+ * Where an attached file ended up, so the agent can be told to read it.
+ *
+ * ── Why a path and not the bytes ──
+ *
+ * An image goes to the model as base64 inside the message. A PDF cannot: it would be megabytes of base64
+ * in the turn, re-sent for the rest of the conversation, to produce text the agent's own tools extract
+ * better. Reading a document is something `Read` already does well — page ranges, and re-reading a
+ * section later without paying for it twice.
+ *
+ * So the file stays where `/upload` put it and the prompt names it. This is a machine where the agent
+ * runs as the person who attached it, so it can simply open it.
+ *
+ * ── The ids are checked, not trusted ──
+ *
+ * An id arrives in a websocket message, and it is used to build a path. `/upload` generates them as
+ * `<uuid>.<ext>`, so anything that is not that shape did not come from us — and a `..` inside one would
+ * name a file outside the attachment directory entirely. Only the two directories uploads actually write
+ * to are searched, and only ids matching the shape uploads actually produce.
+ */
+const ATTACHMENT_ID = /^[a-f0-9-]{36}\.[A-Za-z0-9]{1,8}$/;
+
+export async function attachedFilePaths(
+  userId: number,
+  sessionId: string,
+  attachmentIds: string[] | undefined,
+): Promise<string[]> {
+  if (!attachmentIds?.length) return [];
+
+  const user = await getUserById(userId).catch(() => null);
+  if (!user?.email) return [];
+
+  const paths: string[] = [];
+  for (const id of attachmentIds) {
+    if (!ATTACHMENT_ID.test(id)) continue;
+    // Uploads before a session exists land in `tmp/`; everything after lands under the session. Both are
+    // checked because a first message carries ids from the first shape and later ones from the second.
+    const candidates = [join(getAttachmentsDir(user.email, sessionId), id), join(getTmpAttachmentsDir(user.email), id)];
+    const found = candidates.find((path) => existsSync(path));
+    if (found) paths.push(found);
+  }
+  return paths;
+}

src/servers/api/chat/websocket.ts

@@ -16,6 +16,7 @@ import { sendOpenCodeStreaming } from '@@/channels/send-opencode';
 import * as sidecar from '@@/sidecar-registry';
 import { join } from 'path';
 import { getOwnerHomeDir, getEmailAccountsDir } from '../../../servers/data-path';
+import { attachedFilePaths } from './attached-files';
 import { getUserSettings, getEmailAccounts, getChatEventsSince, appendChatEvent, getUserById } from 'officerdb';
 import { resolveHomeDir } from '@@/user-home';
 import { claudeLoginState } from '@@/os-user-claude';
@@ -360,10 +361,29 @@ async function handleChat(
   const sessionId = msg.sessionId || randomUUID();
 
   // Prepend resume summary to the prompt if present
-  const prompt = msg.resumeSummary
+  const withSummary = msg.resumeSummary
     ? `Here is a summary of a previous conversation to continue from:\n\n${msg.resumeSummary}\n\n---\n\nUser's new message: ${msg.prompt}`
     : msg.prompt;
 
+  /**
+   * Attached files, named by PATH rather than carried in the message.
+   *
+   * `attachmentIds` has been on this wire and in `types.ts` since before today and NOTHING read it —
+   * images work because they travel separately as base64, so the ids were dead. A PDF cannot travel that
+   * way: it would be megabytes of base64 in every turn of the conversation, re-sent forever.
+   *
+   * So the file stays on disk and the agent is told where it is. Reading a document is something its own
+   * tools already do better than a prompt can — page ranges, extraction, re-reading a section later
+   * without paying for it again — and this is a machine where the agent runs as the person who uploaded
+   * the file.
+   */
+  const attached = await attachedFilePaths(userId, sessionId, msg.attachmentIds);
+  const prompt = attached.length
+    ? `${withSummary}\n\nAttached ${attached.length === 1 ? 'file' : 'files'} (read from disk):\n${attached
+        .map((path) => `- ${path}`)
+        .join('\n')}`
+    : withSummary;
+
   // Use provided model, or fall back to user default, or the system default.
   const model = msg.model || (await getUserDefaultModel(userId)) || DEFAULT_MODEL;
 

src/servers/api/upload/upload.ts

@@ -4,6 +4,8 @@ import { join } from 'node:path';
 import { getTmpAttachmentsDir, getAttachmentsDir } from '@@/data-path';
 
 const MAX_FILE_SIZE = 5 * 1024 * 1024; // 5 MB
+/** PDFs never enter a prompt whole — the agent reads them from disk — so the ceiling is storage, not context. */
+const PDF_MAX_FILE_SIZE = 25 * 1024 * 1024; // 25 MB
 
 export const uploadRouter = createRouter();
 
@@ -18,12 +20,21 @@ uploadRouter.post('/', async (ctx) => {
     return ctx.json({ error: 'file is required' }, 400);
   }
 
-  if (!file.type.startsWith('image/')) {
-    return ctx.json({ error: 'Only image files are allowed' }, 400);
+  // Images and PDFs. The two are handled very differently downstream and it is worth knowing which is
+  // which here: an image travels to the model as base64 in the message, while a PDF is left on disk and
+  // the agent is told where it is, because reading a document is something its own tools already do well
+  // (page ranges, extraction) and a 10 MB base64 blob in a prompt is not.
+  const isImage = file.type.startsWith('image/');
+  const isPdf = file.type === 'application/pdf' || file.name.toLowerCase().endsWith('.pdf');
+  if (!isImage && !isPdf) {
+    return ctx.json({ error: 'Only images and PDFs are allowed' }, 400);
   }
 
-  if (file.size > MAX_FILE_SIZE) {
-    return ctx.json({ error: 'File exceeds 5 MB limit' }, 400);
+  // A PDF is allowed to be bigger: 5 MB is a generous photo and a small report. The agent reads it from
+  // disk in page ranges, so size costs storage rather than context.
+  const limit = isPdf ? PDF_MAX_FILE_SIZE : MAX_FILE_SIZE;
+  if (file.size > limit) {
+    return ctx.json({ error: `File exceeds ${Math.round(limit / 1024 / 1024)} MB limit` }, 400);
   }
 
   try {
@@ -41,10 +52,16 @@ uploadRouter.post('/', async (ctx) => {
     await mkdir(saveDir, { recursive: true });
     await Bun.write(join(saveDir, filename), buffer);
 
-    const base64 = buffer.toString('base64');
-    const dataUrl = `data:${file.type};base64,${base64}`;
+    // Only an image gets a data URL. Building one for a PDF would mean carrying megabytes through the
+    // browser for a preview nothing renders.
+    const dataUrl = isImage ? `data:${file.type};base64,${buffer.toString('base64')}` : '';
 
-    return ctx.json({ filename: file.name, dataUrl, attachmentId: filename });
+    return ctx.json({
+      filename: file.name,
+      dataUrl,
+      attachmentId: filename,
+      kind: isPdf ? 'pdf' : 'image',
+    });
   } catch (err) {
     const message = err instanceof Error ? err.message : 'Upload failed';
     return ctx.json({ error: message }, 500);