Hello! Today we launched Retool AI, a set of AI building blocks to help you create AI-powered apps and workflows in minutes.
- Build apps and workflows powered by AI: Use pre-built AI actions to summarize text, classify images, extract entities, and more. OpenAI’s gpt-4 & gpt-3.5-turbo are available out of the box. You can enter your API key, and swap in other LLMs any time.
- Create data-aware AI chatbots in 30 seconds: Upload PDFs, crawl websites, and connect SaaS tools into Retool Vectors, our all-new managed vector store. Ask questions about your data & get helpful answers instantly.
- Code, query, and debug faster: Ask our AI assistant questions, and it’ll generate code that just works, with context into your app’s components and data sources.
Get started today by logging into your cloud account! Retool AI for self-hosted customers will come in the next release—v3.12.
Learn more in the docs: Retool AI | Retool Docs
Questions as you get started? Post in the App Building category or Workflows category of our forum.
The new vectorised database is incredible.
Would love to know how it's attached to the chat system - at the moment you can edit the chat message and system prompt, would be cool to see how the vectorised data is used and what is the process flow of the vector search is.
I am looking forward to Retool AI for self-hosted, any planned release date for v3.12?
this is great @nate! The vector DB is crazy - So happy you guys went further than just AI helper bots. Retool making ACTUALLY useful AI features.
@nate Following on from the OpenAi announcements today, can you tell us what plans Retool has to integrate these features and a possible timeline?
Hey @maillme thanks for the question. We work closely with OpenAI so that we can support their latest updates in Retool with our native integration. Happy to share that we support GPT-4 Turbo (with vision) and DALL·E 3 in Retool today.
Thank you so much for the response, really glad to see that this is already integrated!
@nate - sorry if this is a really silly question (As I'm sure the name we see isn't the actual model), but I need to ask.......
Is Retool defo referencing the correct vision model? As I believe it should be appended with: -preview
I ask because I am getting the following error - and I belive my OpenAi account shoudl be able to access:
FYI - I accessed the gpt4-vision-preview API manually through Retool, and it works - but, it only works with a url, and not with Base64 data.
When I try with GPT manually, it says the Base 64 data is too long.
So, my question to you is - has anyone in Retool managed to get GPT4-vision to provide a response with a Base64 file (I.e. file stored on retool storage) ?
Also - my previous question still stands, I think it is maybe incorrect (the API) in your end. As I can do manually myself.
thanks a lot for any help!
This is a big use-case for us if we can get GPT vision working through retool.
Hey @maillme - which AI Action type are you using? I just tried building an app with the "Generate text from image" action type (with gpt-4-visual selected as the model) and it seems to work for me with base64 data. Can you send over a screenshot of your setup? Alternatively, feel free to email me directly: firstname.lastname@example.org
Any ETA when GPT-4-Turbo will be available for the Chat Generation use in the Retool AI module? Notice that the default 4 models are the only options available at the moment for GPT4/32k and GPT3.5/16K.
Thanks @dworsky for the reply,
I've sent you an email on this.
I tried to recreate your example - but still get the same error I was getting previously.
So to test my API Key etc (which I had tested outrwith Retool already) - i created an API resource, and ran a GPT vision query manually - and this seemed to work (using a URL, as the Base64 data is too large for some reason).
So I think my API Key is fine.
So I'm at a loss - check your inbox,
thanks again - really appreciate the help here,