Chaindesk - open-source ChatGPT Agent for your data and website

[marcusquinn]

• https://www.chaindesk.ai/ (formerly Databerry https://www.databerry.ai )
• https://github.com/gmpetrov/databerry
• https://twitter.com/georges_petrov/status/1650488414100238337

Build Your Own
ChatGPT Agent
Trained On
Your Custom Data
With our no-code platform, you can create a custom AI chatbot trained on your data in seconds. Streamline customer support, onboard new team members, and more!

Efficient Your AI chatbot can answer frequently asked questions and handle simple support requests, allowing your team to focus on providing more personalized assistance to customers.
24/7 Support Your AI chatbot can handle customer inquiries around the clock, providing efficient support even when your team is unavailable.
Easy to Implement Integrating your AI chatbot onto your website is a breeze - simply copy and paste our code onto your site and start providing instant support to your visitors.

[robi]

@marcusquinn great find!

I architected something similar to this while I was at IBM to solve many of the pain points I was seeing.

They rejected my solution.

This was in 2017.

I left soon after.

[LoudLemur]

@robi I am currently looking around to find an AI solution like this for a production environment.

Basically, they receive a lot of electronic communications and it would help free staff time for other work if the AI could handle a lot of the emails and maybe run a chat, too.

The AI would ideally not have any telemetry with e.g. OpenAI.

At the moment, I am wondering how to form an in-house "knowledge base" on which to train the AI. For example, training it on the website would be relatively easy, but how would one go about converting one's internal or external communications like emails into training material? It is also challenging to work out how to on-board the staff so that they would start setting up the AI with the data it needs.

If anybody knows a place where people have some real-life stories about how they went about this and what happened, I would like to know.

[robi]

@LoudLemur that's literally what this project does.

3 weeks ago, Georges made a video on this, many others exist too:

Pay attention to what and how he does it.

It's not so useful to "convert" all internal data as it's out of context with each other, and you'd need to retrain the whole LLM on all that.

If you use an offline private pretrained LLM, it's better to prime it with the context and other prompts before the customer gets to query it.

Then stitch together the various parts in relevant contexts (docs, support case email thread, self support on all cases, faq, website, etc.)

It could all be the same chat UI window, but different contexts on different pages, or opt for one inclusive chat UI with internal routing via links or personas, ...)

If you're not into doing some dev work, wait a bit for things to develop as its' moving fast.

Influence the creation of what you need. Keep an eye on the relevant Github repos and participate with the creators suggesting features and other projects that have useful parts to integrate making what you want that much closer to reality.

[LoudLemur]

@robi said in Databerry - open-source ChatGPT Agent for your data and website:

@LoudLemur that's literally what this project does.

3 weeks ago, Georges made a video on this, many others exist too:

Pay attention to what and how he does it.

It's not so useful to "convert" all internal data as it's out of context with each other, and you'd need to retrain the whole LLM on all that.

If you use an offline private pretrained LLM, it's better to prime it with the context and other prompts before the customer gets to query it.

Then stitch together the various parts in relevant contexts (docs, support case email thread, self support on all cases, faq, website, etc.)

It could all be the same chat UI window, but different contexts on different pages, or opt for one inclusive chat UI with internal routing via links or personas, ...)

If you're not into doing some dev work, wait a bit for things to develop as its' moving fast.

Influence the creation of what you need. Keep an eye on the relevant Github repos and participate with the creators suggesting features and other projects that have useful parts to integrate making what you want that much closer to reality.

Excellent advice. Thank you very much. Having a basic AI and then tweaking it with some data from the organization would probably be the way to go.

[LoudLemur]

@marcusquinn OpenAI is rolling out plugins for all ChatGPT Plus users this week, so DataBerry will have a chance to shine.

Has anybody tried using it?

[marcusquinn]

@LoudLemur I'm getting great results with the Meow AI Engine plugin for Wordpress. Since that's solving my immediate needs elegantly, not looked any further at any others, but imagine they all work on the same principles, but need their own hosting, where Wordpress is already somewhat of GPL app operating system.

[robi]

@marcusquinn said in Databerry - open-source ChatGPT Agent for your data and website:

Meow AI Engine

Magical Meow indeed.

[LoudLemur]

This incredible software seems to now be called Chaindesk:
https://www.chaindesk.ai/

Maybe @marcusquinn can update the thread title?

There is currently an Application request for Apache Answer on Cloudron. I wonder whether AI tools like this might make Apache's Answer redundant, or whether they would integrate well together. You might even be able to populate an Apache Answer instance using a Chaindesk AI Agent.

By the way, check out the new ChatGPT4 Turbo, which has a huge 128K tokens context (relative to the previous 4K), which is available on Chaindesk now.

[marcusquinn]

@LoudLemur Updated, ta.