Mrs Doubtfire explains AI – Fabrk.ai build custom AI agents quickly

Hacker News

Share card

Mrs Doubtfire explains AI – Fabrk.ai build custom AI agents quickly

TLDR - Launching Fabrk.ai - No code solution for building AI agents Thanks for checking in. Sorry for the clickbait title. I created a Mrs Doubtfire agent and asked it to explain Fabrk - see below. Live examples using Fabrk: - Embed demo: https://input.fabrk.ai/c/fabrk-demo/fabrk-demo - https://www.tryplenti.com/ - https://plantprovisions.com/ - SMS integration: https://nativepet.plentiai.com/try/native-pet?campaignId=d1e... I’ve been building on top of Open AI’s api since the launch of GPT 3.5. What I’ve realized is that building production ready agents is hard. Problems - API integration - You need a savvy dev to integrate with the Open AI api to do anything meaningful. This seems obvious but several companies have built demos with Open AI’s GUI but couldn’t ultimately get the demo to production. - Context is hard - It is hard to get the necessary data into the context at the right time. Companies don’t understand the need for RAG. Any meaningful app needs to pass data in and get data out of the chat. - The APIs are changing rapidly. We started using Open AI’s completions api and have switched to the assistants api. There are important differences that affect what you can build. - Odd things happen in production. This is par for the course for anyone who builds software. But the non-deterministic nature of the apis seems to exacerbate the edge cases. Or at least make them seem more dramatic. - Need introspection. For any agent that uses tools or complex context you need a way to understand why the agent responded the way it did. Did it pull in the wrong context? Did it not call it a tool when it should have? Was it just a hallucination? You need a way to introspect the chats. - Using AI in different places. Companies want to use the chats in different mediums - sms, email, embedded on their website or dashboards, etc. How Fabrk works - No code solution for building agents. We have a GUI to build the standard file_search and data_analysis agents. But we have many other other agents - location search, shopify, api calls, storing memories and sending money. We are building more daily. Our plan is to let anyone create and share agents. - Passing data in and out of the chats can be done with a couple clicks. Upload files to get data in. Add Data Extraction to get data out. - We have a front row seat to the shifting api surface area. We know where the gotchas are. You don’t have to hire a team to build your app. - We can provide white glove service to help get your agent in production and provide ongoing support to fix issues. - Our GUI provides introspection. You can see how the AI uses tools and context to come up with a response. This lets you iterate quickly. We also have a dashboard that lets you see all the chats your users are having with the agent. - Out of the box we support an embedded chat app, text message or email chatbots. Feel free to ping me if you want a demo or have questions about Fabrk. As Mrs Doubtfire says - Keep knitting those dreams. _ Oh, hello dearies! Gather 'round your digital fireplace, 'cause Granny's got some exciting news for you! Would you believe it if I told you that creating custom AI chatbots is now as easy as knitting a cozy scarf? Here are some of the fabulous features Fabrk offers: Create and customize agents for your chats. Store files for reuse (useful for all those patterns!). A UI to show the context and agents used in responses. Data extraction tools to glean customer insights. Use it in your browser, through email, or SMS/text. Embedding tools to seamlessly add chatbots to your website. Okay, dearies, I'll stop spinning my yarn here. Whether you’re an author, an influencer, or run a healthcare app, Fabrk is here to make your life easier. Pop on over to Fabrk and see how this lovely tool can help you! Keep knitting those dreams, Mrs. Doubtfire