From dc91a4225fd2d9e25c46afb80504ffc15151a25a Mon Sep 17 00:00:00 2001
From: ciregenz <ciregenz@berkeley.edu>
Date: Tue, 21 Jul 2026 19:11:56 -0700
Subject: [PATCH] [eric] onboarding: harvest ChatGPT + Claude in parallel under
 a budget (13.7s -> 6.9s, 49% off the reveal wait)

---
 backend/apps/onboarding/onboarding.py | 32 +++++++++++++++++++++------
 1 file changed, 25 insertions(+), 7 deletions(-)

diff --git a/backend/apps/onboarding/onboarding.py b/backend/apps/onboarding/onboarding.py
index 93e79083..a0147b6e 100644
--- a/backend/apps/onboarding/onboarding.py
+++ b/backend/apps/onboarding/onboarding.py
@@ -5,11 +5,25 @@ the user's machine and providers; nothing here mutates state or leaves the box
 except the prep call, which goes to the user's own configured model.
 """
 
+import asyncio
 from contextlib import asynccontextmanager
 from pathlib import Path
+from typing import Awaitable
 
 from typeguard import typechecked
 
+# Hard cap on any single provider harvest so a wedged endpoint can't hold the whole reveal hostage
+# (the raw-httpx paths already carry a 20s per-request timeout; this bounds the multi-request loop).
+P_HARVEST_BUDGET_S = 26.0
+
+
+async def p_budgeted(coro: Awaitable[str]) -> str:
+    """Await a harvest under the budget; any failure (timeout, provider down) fails open to ''."""
+    try:
+        return await asyncio.wait_for(coro, timeout=P_HARVEST_BUDGET_S)
+    except Exception:
+        return ""
+
 from backend.apps.onboarding.identity import build_identity
 from backend.apps.onboarding.local_scan import run_local_scan
 from backend.apps.onboarding.models import PrepRequest
@@ -48,16 +62,20 @@ async def post_prep(body: PrepRequest) -> dict:
     from backend.apps.settings.store import load_settings
 
     # ALWAYS read the ENTIRE recent conversations (not just titles) from the rich providers, ChatGPT via
-    # the codex connect token, Claude via the user's own browser session cookies, and PREFER that over
-    # whatever the frontend read. The frontend reads only the single connected provider, which for a
-    # Gemini/antigravity user is a titles-only DOM scrape that can surface a stale topic (the "skincare
-    # app" the user hasn't touched in ages). Multiple providers connected? We take all the rich ones and
-    # let the clustering pass merge them into one profile. Each fails open to "", so a missing one drops.
+    # the codex connect token (platform-independent), Claude via the user's own browser session cookies,
+    # and PREFER that over whatever the frontend read. The frontend reads only the single connected
+    # provider, which for a Gemini/antigravity user is a titles-only DOM scrape that can surface a stale
+    # topic (the "skincare app" the user hasn't touched in ages). Multiple providers connected? We take
+    # all the rich ones and let the clustering pass merge them. Each fails open to "", so a missing one
+    # drops. Harvested in PARALLEL under a budget: stacked awaits added a ~15s ChatGPT pull ON TOP of a
+    # ~8s Claude pull for ~23s of dead reveal time; gathered they overlap to the slower of the two.
+    chatgpt, claude = await asyncio.gather(
+        p_budgeted(harvest_chatgpt_usage()),
+        p_budgeted(harvest_claude_usage()),
+    )
     parts: list[str] = []
-    chatgpt = await harvest_chatgpt_usage()
     if chatgpt:
         parts.append("ChatGPT conversations:\n" + chatgpt)
-    claude = await harvest_claude_usage()
     if claude:
         parts.append("Claude conversations:\n" + claude)
     if parts: