Rivet, open-source AI agent environment by Ironclad Research

[Hyperlooper]

What sets this tool apart from langflow and flowise is you included logic and flow control functionality. That makes so much sense to include because node based workflows really lend themselves to that sort of thing so you really give up a lot when they aren't part of the toolchain.

[HarpaAI]

🎯 Key Takeaways for quick navigation:

00:00 🤖 *Introduction to Rivet, an open-source AI agent environment*
- Rivet is an open-source visual AI programming environment developed by Ironclad Research.
- Rivet was created to build complex AI agents using chains of LLM prompts.
- It allows for real-time observation of AI agent responses and debugging.
00:28 🛠️ *Debugging with Rivet and remote debugger*
- Rivet's remote debugger enables real-time observation of GPT 3.5 responses.
- This feature helps identify areas where the AI agent is failing or brittle, facilitating quick iterations.
01:27 🔄 *Dynamic external execution with Rivet*
- Rivet allows external execution, such as running calculations programmatically.
- It permits the dynamic definition of functions based on user permissions.
02:55 ➕ *Playing the game of 24 with Rivet*
- Rivet demonstrates how to teach an AI agent to play the game of 24 by combining numbers with arithmetic operations.
- Users can edit prompts and sub-graphs directly in Rivet, facilitating experimentation.
04:21 🧩 *Using regex and external calls for refining responses*
- Rivet shows how to use regular expressions and external calls to extract specific information from AI agent responses.
- This allows for fine-tuning and refining the AI agent's output.
05:20 🔄 *Transitioning to GPT-4 and advanced capabilities*
- The video concludes by mentioning the possibility of transitioning to GPT-4 for improved results.
- It highlights advanced capabilities like comparison formulas and iterative approaches in Rivet.

Made with HARPA AI

[vbywrde]

This is utterly fantastic. Thank you for making this open source! It's hugely appreciated. Wonderful job programming it as well. Brilliant. Just have one question: do you have a plan to allow it to be used by local models, such those on hugging face? While I adore the functionality presented here, my business leaders are concerned about the use of OpenAI for a variety of reasons, and much prefer that we focus on local model solutions wherever possible. Anyway thanks so much! Fantastic!

[ackiamm]

Thanks Sir. I love the node-base interface

[ParaZen]

Slick tool! Thank you so much for this new toy

[EddADP]

Great tool. Way more intuitive than langchain for non developers. Would really appreciate if you could add some more documentation/training vids. I feel like I am just scratching the surface here.. For example can I connect to APIs like SERP for search? or use a local Vector DB like Chroma? Thank you once again for sharing this..

[muhammadahsan]

Thanks for the video, I'm a Python dev so just curious is there any option for not using Node.

[juancarlosbaezpozos]

Summary: Forking flowiseai and let's rename to "rivet", that's all !

[shivanidwivedi]

i had made chatbot using rivet now how can i use it like you are using with localhost

[shemkipruto]

Windows defender strongly prevents installation of this, is there something wrong

[patrick-spacetimelayers]

What a strange video. Is this the launch vid& promo for website?
You don't let a language model calculate, you use a Wolram API.
Switching to GPT4 to solve this really made me question my enthousiasm.