Document piles
Run a folder of PDFs or CSVs through a model without a single line item on an invoice.
Drive ChatGPT and Gemini from Python, for free.
No API keys, no billing — just the free web UI. Send prompts, attach files, and pull back text or images, all from a script.
See it work
ChatGPT Gemini
no api key · nothing to bill
hermex.query("")
The whole idea, in twenty-five seconds.
What it does, why it exists, and a few things people use it for.
From a script to an answer, automatically.
your script
types it in
the real web app
reads it back
text and images
Hermex opens a real Chrome window, signs in with your own session, types the prompt, uploads your files, waits for the reply, and hands it back as a Python object. No API key anywhere in that chain.
One call. Any task.
The same fluent call handles text, files, and images. Flip the switch — only the import and the class name move.
ask.py
from hermex import ChatGPT
res = ChatGPT.simple_query(
"Explain quantum tunneling simply"
)
print(res.text)
from hermex import ChatGPT
res = ChatGPT.simple_query(
"What does this receipt say?",
attachments=["receipt.jpg"],
)
print(res.text)
from hermex import ChatGPT
res = ChatGPT.simple_query(
"Summarize this in five bullets",
attachments=["report.pdf"],
)
print(res.text)
from hermex import ChatGPT
res = ChatGPT.simple_query(
"Extract all the text here",
attachments=["screen.png"],
)
print(res.text)
from hermex import ChatGPT
res = ChatGPT.simple_query(
"A red fox in loose watercolor"
)
res.image.save("art.png")
from hermex import ChatGPT
res = ChatGPT.simple_query(
"Describe what is happening here",
attachments=["photo.jpg"],
)
print(res.text)
from hermex import ChatGPT
res = ChatGPT.simple_query(
"What trend does this chart show?",
attachments=["chart.png"],
)
print(res.text)
from hermex import ChatGPT
res = ChatGPT.simple_query(
"Translate this to French",
attachments=["memo.txt"],
)
print(res.text)
from pathlib import Path
from hermex import ChatGPT
for path in Path("reviews").glob("*.txt"):
res = ChatGPT.simple_query(
f"Sentiment: {path.read_text()}"
)
print(path.name, res.text)
output
$ python ask.py
Quantum tunneling is when a particle slips
through a barrier it shouldn’t have enough
energy to cross…
$ python ocr.py
BLUE BOTTLE COFFEE
1× Cold Brew 4.50
1× Almond Croissant 3.75
TOTAL 8.25
$ python summarize.py
• Revenue up 12% quarter over quarter
• Churn flat at 2.1%
• Hiring paused until Q3
• Two enterprise deals slipped…
$ python extract.py
Settings › Billing
Current plan: Pro — $20 / month
Next invoice: Nov 4, 2026
art.png · 1024×1024 · saved
$ python describe.py
Two people share one umbrella on a
rain-slick crosswalk at dusk, headlights
smearing across the wet asphalt.
$ python chart.py
Steady growth through Q2, then a sharp
dip in July — the shape looks seasonal
rather than structural.
$ python translate.py
Objet : mise à jour du planning
L’équipe livrera la version 2.1 avant
la fin du mois…
$ python batch.py
r-001.txt positive
r-002.txt negative
r-003.txt positive
r-004.txt neutral
→ Returns A plain-text answer — the model’s reply as a string you can print, log, or pass along. The receipt’s full text — every line item transcribed into a string you can parse or save. A short summary — the whole document distilled into a handful of bullet points. The extracted text — everything readable in the image, as one plain string. A generated image — handed back as an object and written straight to the path you choose. A caption — a natural-language description of what is actually in the picture. The takeaway — the trend the chart shows, explained in words instead of pixels. The translated text — your document rendered in the target language, formatting intact. One answer per file — loop a directory and collect a result for every item in it.
Built for the small stuff that adds up.
Internal tools
Small scripts for you or your team, where standing up metered API billing isn't worth it.
Prototyping
Try the idea first. Move to paid API infrastructure once you know it's worth paying for.
One-off asks
OCR a receipt, read a chart, caption an image — the errands the official API is overkill for.
Why pay per token for what the browser already does?
ChatGPT and Gemini are extraordinary. But for OCR, a quick question, or a one-off image, standing up an API key, a billing account, and a cost dashboard is a lot of ceremony. Hermex hands you the same models — minus the second invoice.
the usual route
Official API
- An API key and a billing account to set up first
- Metered — every token is a line item
- Separate rate limits and usage quotas to track
- A second bill that grows with every run
this library
Hermex
- No API keys, no billing setup, nothing to rotate
- The same ChatGPT and Gemini models you already use
- Runs on your existing login — free tier included
- One fluent interface for both platforms, files and all
Questions asked frequently.
Q: Do I need an API key?
No. Hermex drives the ordinary ChatGPT and Gemini web apps — the same ones in your browser right now. There is no key to create, store, rotate, or pay for.
Q: Is it really free?
Hermex itself is free and open source under MIT. It runs on the ChatGPT or Gemini access you already have — including the free tiers — so there's no separate metered bill. It doesn't unlock anything your account can't already do.
Q: How does it work under the hood?
It launches a real Chrome window with
undetected-chromedriver, types your message the way a person would, uploads any attachments, waits for the reply to finish streaming, and hands it back as a Python object. The browser profile persists on disk, so your login carries across runs.Q: Do I have to log in?
Not for plain text questions — those work in guest mode. Logging in unlocks the rest: file uploads on Gemini, image generation, and anything else tied to your account.
Q: Which models are supported?
ChatGPT and Google Gemini today, behind one shared interface. Swapping between them is a single class name —
ChatGPT to Gemini — and nothing else in your script changes.Q: What can I attach?
Images, PDFs, CSVs, JSON, and plain text — the same things you'd drag into the chat window by hand, passed as a list of paths.
Q: Is this meant for production?
Honestly, no. It drives a real browser, which makes it slower and more brittle than an official API, and a UI change upstream can break it. Hermex is built for scripts, research, and small automations — not high-volume production traffic. Each platform's Terms of Service and rate limits still apply to how you use it.
Start in one line.
Free and open source, running on the web apps you already use.
pip install hermex