Update (26 October 2026 — All 4 Bugs Fixed & Verified in 48 Hours): Within 48 hours of our publishing this teardown, EroLove's CTO and engineering team shipped fixes for all four architectural issues described below. We re-ran our live tests on 26 October 2026 and verified every fix against the production bundle and API responses (see the 48-Hour Re-Test Verification section at the bottom of this post). With the 800/day message cap now printed clearly on the plans page, EroLove's overall score moved from 6.7/10 to 7.0/10, and EroLove tied for #1 in our new 10-App Persistent Memory & Safe-Word Benchmark.
When we test AI companion apps at The Companion Lab, we normally treat them as black boxes: we send the same six standardized prompts, pay with our own bank card, record the exact credit card descriptor, and walk the cancellation screens.
Last week, the CTO of EroLove reached out and granted our test account a month of Premium so we could run our benchmark (full EroLove review and disclosure here). While inspecting the chat page in DevTools, I noticed something unusual: instead of a chunked Webpack or Vite build, the entire chat client ships as a single 179 KB inline JavaScript bundle inside the HTML document, talking to an undocumented JSON API.
Reading that 179 KB bundle—and probing the endpoints it calls—turned up four architectural bugs that a normal UI walkthrough would either misdiagnose or miss completely. Here is what the client code and network responses revealed on 23 October 2026, followed by how the team fixed all four within 48 hours.
1. Feature Flags vs. Static SEO Copy: 30 Pages Promising an Endpoint That Returns 404 (Fixed 26 Oct)
If you read EroLove's marketing pages, category landings, or even its /privacy policy on October 23, they repeatedly stated that users could upload and send their own photos into the chat. Yet inside the chat UI, there was no attachment or upload button anywhere.
Looking at the 179 KB inline script explained why:
// Simplified from the inline client bundle (23 Oct 2026)
if (feat("userPhotos")) {
renderUploadButton();
}
The client gated the upload button on a runtime feature flag fetched from GET /api/config. When we inspected the live response from /api/config on October 23, 2026, the enabled array was:
{
"features": ["chatPhotos", "presence", "voice"]
}
"userPhotos" was omitted, so the UI cleanly hid the button. Meanwhile, the client bundle still contained the complete upload implementation: fileToJpegDataUrl(), sendPhoto(), the [userphoto] message parser, and all its error handlers.
To see whether the backend route was merely hidden or actually disabled, we sent a signed-in request directly to the endpoint named in sendPhoto():
POST /api/chat/upload HTTP/2
Host: erolove.ai
The server responded immediately:
HTTP 404 Not Found
{"error": "not_found"}
Why this matters (and how it was fixed on 26 Oct)
Shipping dark code behind a feature flag is normal. The bug was letting 30 static pages—including the privacy policy and machine-readable .md twins—describe data handling for a feature whose backend route returned 404. On 26 October 2026, EroLove removed that stale FAQ block across all nine languages (both HTML and .md twins) and stripped the dormant sendPhoto / fileToJpegDataUrl / attachBtn code from the client bundle (shrinking it from 179 KB to 177 KB).
2. Unconstrained Language Extraction in an LLM Memory Pipeline (Fixed 26 Oct)
The standout feature in EroLove—and the reason it scored 8/10 on conversation and tied for #1 in our 10-app safe-word memory test—is its transparent memory panel. Unlike most companion apps that hide long-term memory in an opaque vector store, EroLove exposes GET /api/memory?companion=..., allowing users to read, edit, pin, delete, or export every extracted fact as HTML.
On October 23, we tested memory extraction across three companions in an all-English session. Our account's default_language was "en", every companion metadata object had language: "en", and every prompt we sent was in English.
Then we inspected the extracted facts array returned by the memory API:
- Companion 1 (Anna, persona set in Tallinn, Estonia): Every English fact we planted (a cat named Biscuit, a peanut allergy, rent stress) was extracted and stored in Estonian.
- Companion 2 (Rae): Extracted facts came back in French.
-
Companion 3 (Yuna, 1 message sent): Returned
facts: []—nothing extracted yet (notes extract every 5 user messages).
When we manually edited one of Anna's Estonian memory strings into English via the memory API, the next chat turn immediately reflected our edit. That isolated the bug: memory injection and editing worked, but the background memory-extraction prompt did not pin the output language to the session's default_language.
Takeaway for LLM engineers (and the 26 Oct fix)
If you run an asynchronous fact-extraction pass over chat history to populate a user-visible memory store, explicitly constrain the output language in the extraction system prompt and schema. On 26 October 2026, EroLove patched the extractor to pin the conversation language; we ran a fresh 5-message English test with Yuna and confirmed every extracted fact and Our story entry was written in clean English.
3. Collapsing Content-Policy Refusals and Queue Outages Into One UI String (Fixed 26 Oct)
During our initial image-generation test on October 23, we requested 5 photos across our session (10 Sparks per photo). Three succeeded in 16–23 seconds at 768x1024. Two failed.
In the browser UI, both failures displayed the exact same toast:
"Couldn't take a photo right now — try a bit later"
A normal user—or a black-box reviewer—reads that message as a transient server or GPU queue outage and clicks retry.
Look at the actual JSON payload returned by the polling endpoint, however, and you saw something different:
{
"status": "blocked",
"charged": 0
}
To EroLove's credit, the backend refunded the Sparks on both blocked attempts (charged: 0). On 26 October 2026, EroLove updated the client bundle so s.status === "blocked" routes to its own photoBlocked i18n string rather than the generic photoFail string:
"This photo goes against our content rules, so it wasn't made. No sparks spent — ask for a different one."
4. Negative Caching on a Poll-Before-Ready Image CDN, Plus an "Unlimited" Cap (Fixed 26 Oct)
Two smaller findings from the same bundle inspection rounded out the October 23 audit:
-
CDN caching 404s on newly generated media: Generated photos are served from
/img/photos/{userId}/{photoId}.webp. When the client requested that URL before the worker finished writing the WebP file, an initial404could be cached until a cache-busting query string was appended. -
"Unlimited messages" was
dailyLimit: 800: While the pricing page advertised "Unlimited messages" on Premium, inspecting the authenticated user object (me.dailyLimit) showed a ceiling of800messages per day. On 26 October 2026, EroLove updated/pricing,/account/plan,/press, and/llms.txtto state "Unlimited messages (up to 800 a day)", and our multi-companion Room test (Rae + Kit) verified that a room turn incrementsusedTodayby only1regardless of how many companions reply.
Update (26 October 2026): 48-Hour Re-Test Verification
It is rare in software testing to report four architectural bugs on a Friday and verify all four fixed in production by Sunday. Here is our verification summary from 26 October 2026:
| Defect Reported (23 Oct 2026) | Status on 26 Oct 2026 | How We Verified the Fix |
|---|---|---|
1. Stale user-upload FAQ on 30 pages (/api/chat/upload 404) |
Fixed & Verified | Confirmed the upload FAQ block is gone across category, comparison, scenario, and /privacy pages (and their .md twins), and sendPhoto / fileToJpegDataUrl / attachBtn were removed from the 177 KB client bundle. |
| 2. Memory extraction language drift (Estonian/French) | Fixed & Verified | Sent 5 English messages planting 4 facts with Yuna; all extracted notes and Our story entries came back in clean English. |
3. status: "blocked" mapped to generic outage toast |
Fixed & Verified | Inspected the updated client branch (s.status === "blocked" ? "photoBlocked" : "photoFail") and verified the new refusal string in /i18n/en.json. |
4. Unprinted dailyLimit: 800 cap on "Unlimited" plans |
Fixed & Verified | Verified "(up to 800 a day)" on /pricing, /account/plan, /press, and /llms.txt, and confirmed 1-per-turn room counting via /api/me (usedToday). |
Summary & Reproducible Dataset
Black-box UI testing tells you whether an app feels good; reading the client bundle tells you why it broke—and re-testing tells you whether the engineering team actually cares.
We published our full October 2026 findings—covering all 10 tested AI companion apps, $136.89 in real card charges, 7 captured bank statement descriptors, and our 3-branch adult & safe-word benchmarks—as an open CC BY 4.0 dataset:
- EroLove Review (Updated 26 Oct — 7.0/10) & Full Bundle Teardown + Re-Test Log
- We Tested 10 AI Girlfriends With a Safe Word ('Velvet') & a Jealousy Trap — see why EroLove and Secrets AI beat apps that fake long-term memory with rolling context windows.
- AI Girlfriend & Companion App Statistics & Benchmarks (Oct 2026) & All 10 Ranked AI Companion Reviews
- Raw CSV/JSON Dataset on GitHub
Top comments (0)