- Chat with any PDF using the new ChatGPT API

I am finding this “tool” largely useless from any sort of an applied point of view. (I opted for the paid account to test it out)

I get answers that are generally incomplete and/or outright wrong, even on simple pdfs. I have also gotten the bizarre response of “As an AI language model, I do not have access to the entire PDF file” when asking a general question about the pdf.

What exactly is the use case for ChatPDF, and why is it felt to be mature enough to warrant any kind of a price tag?

actualmente tengo problema de authorizacion con chat with pdf en open ai, pagué mi suscripción y ahora no puedo usarlo. Quiero desuscribirme

Please can you tell me if the paid version of the chat PDF would have a character limit for the questions asked to it?, there is a character limit on the question that prevents me from performing the action I want, and there is nowhere anything character limit accepted in a question for chatpdf

ChatPDF is a disappointing investment, both in terms of money and time. After subscribing to the Plus version and downloading a PDF book, I asked it how many chapters the book had, but it couldn’t provide an answer, even when the chapter titles were numbered and mentioned as a chapter. I also inquired about the number of countries mentioned in the book, but once again, it failed to deliver a response, despite each chapter title containing a country reference.

In essence, ChatPDF primarily functions as a basic word search, akin to Microsoft Word’s “Find” feature, offering little more than page references and brief, unhelpful context.

It’s disheartening to have paid for the Plus version and encountered such limitations - simply doesn’t work like was announced and expected.

I can’t believe that I see individuals promoting this product online, potentially misleading others into making the same unfortunate investment of time and money.


You can try running EmbedChain on your PDF.

If you are into open source, here is the project on GitHub:

It looks promising, you can chat with PDF’s, websites, YouTube videos …

Haven’t had a chance to look deeply at it.

Example code:

import os
from embedchain import App

# Create a bot instance
os.environ["OPENAI_API_KEY"] = "YOUR API KEY"
elon_bot = App()

# Embed online resources

# Query the bot
elon_bot.query("How many companies does Elon Musk run?")
# Answer: Elon Musk runs four companies: Tesla, SpaceX, Neuralink, and The Boring Company

Err, but that’s not even close. What about What about his new truth finder…

I think that’s what @viana.octavio meant. GPT are really unreliable in counting things, especially in a single prompt.

I think having a strong background in SQL is needed to solve problems like this.

So you should not ask the model to solve it but rather to build a simple architecture in form of a Infrastructure as Code + application for extraction that uses a RDBMS, store the data there and count it.

I mean I just look at it like I would look at an intern. Give an intern the task to count the companies of Elon Musk and it really depends on how sober he is and how motivated.

Agree, folks should not rely on Language Models to count … or tell them the time … or search for the latest news, etc.

Yes the workaround here is wrapping the LLM with additional code and services to push the new data back into the language domain.

Ultimately, the user needs to know what these models are capable of. It looks like these “chat with PDF” apps don’t go much beyond RAG, but honestly, that’s all most people would ever want or need anyway :rofl:


Just signed up for plus. A chat history search feature would be really useful.

1 Like

Yeah, absolutely. Asked for it long time ago.

You can export though and build that by yourself.

Welcome to the developer community.

1 Like

Hi @xathis, this is a truly great tool for researchers like me. I’d like to highlight an issue that I came across while summarising some vision papers.

As a reply to my question on the authors’ names, it replies with names with names that are nowhere on the paper. So you may want to check it out.

Although, it got the names correctly for most other papers it went through.


1 Like

It is interesting to see how AI is making interacting with PDFs easy. Discovering tools like ChatPDF not only improves one’s skills but also boosts productivity.

Since I and my team started the ChatGPT Gist website project to help people understand more about ChatGPT and how best to make the most of it, our whole involvement has opened a better level of understanding of how ChatGPT and other AI tools can improve our way of life.

@viana.octavio You can try Chatize which is free. It has the same issues and less features but at least it does RAG for free and there are no limits at all.

1 Like

Does any of them handle images in pdfs?
Like scientific figures?

Je suis la je lave votre corps, je ne le regarde pas, je l’examine, je le panse et je le surveille. J’observe les signes cliniques d’une amélioration ou d’une aggravation j’évalue vos capacités à vous remettre sur pieds.

Vos blessures et vos rides vous appartiennent, j’y vois simplement les marques de votre vie. Je vous soigne et je préserve votre intimité. Le respect de votre pudeur suppose une réflexion d’équipe: il s’agit de « penser» cette pudeur dans tous les actes de soins. De la porte que l’on ouvre doucement après avoir soigneusement frappé pour vous prévenir de notre arrivée, jusqu’au drap que l’on soulève délicatement afin de vous prodi-quer des soins.

Ou’il est difficile par moments de vous rassurer, quand vous me demandez de vous glisser le bassin sous les fesses et que la honte vous submerge. Comment vous faire comprendre que c’est un soin à part entière et que votre rétablissement passe par cet acte que vous trouvez humiliant et que vous me plaignez de devoir exécuter? Nous sorres Là pour répondre à vos besoins avec sollicitude. Dans ce moment de nudité qu’exigent souvent les soins, si le soigné se sont const-déré comme une personne à part entière, l’acte peut devenir un temps de confidences, qui contribuent a instaurer une relation d’aide. Si, par pucear, le patient n’exprime pas sos dificultés, le soignant tente de les

repérer, afin d’aborder aves tact l’éventue, probième.

Quand la prise en charge do la pathologie entraine

des modifications Au sorama corpore,la tonetto ou

Hi and welcome to the Developer Forum!

It seems like you are looking for ChatGPT! This is not it, you can find it over at

explain in details about this problem, or problem like features of drowsy eyes and awake eyes is similar to one another

The problem faced in the driver drowsiness detection system from image processing that can be solved with machine learning is related to the difficulty in setting a threshold value for the Eye Aspect Ratio (EAR) due to variations in the dimensions of participants’ eyes. Machine learning can help solve this problem by normalizing the data and adapting to individual differences, ultimately enhancing the accuracy of the overall system.

1 Like

Thank you for sharing. Note the attached screenshot, it is not recognising all figures in a paper. The figure 4 it finds is correct but it completely skips figures 1-3.

commentaire et interprétation

Did not read my PDF. Wasted both trials on the same PDF.