Subscribe
Sign in
Rivet by Ironclad — Open-source visual AI programming environment from Ironclad
Rivet is a visual programming environment for building AI agents with LLMs. Iterate on your prompt graphs, then run them directly in your application. Rivet was born out of our need at Ironclad to build complex AI agents within our existing application.
25
Replies
Best
Great work on the launch, Ironclad team! 🥳 Can't wait to use this for more complex generative pipelines.
Curious: what was the motivation for providing a Tauri desktop app for the initial release? Are there workflows present in the desktop version that make development easier?
@vivek_nair thanks!
Actually, when running locally, you can open it in a web app. There's one catch, though: saving project files is pretty janky (it downloads the project file). I think basically everything else works pretty well, though!
Very cool, congrats on the launch Cai and team! Who's the target ICP - mid/late stage enterprises or startups/tinkerers? Or both?
Thanks @damndan!
ICP is probably early to growth-stage start-ups!
That's at least what a lot of our early design partners and collaborators have been. Rivet slots in neatly to most modern TypeScript applications, and is pretty easy to adopt progressively.
(Also depends on your definition... I'd consider Ironclad a "growth-stage start-up," but I imagine others would say otherwise!)
What is the maintenance plan for this? I see that you're primarily working on the Ironclad contract AI. Would be curious if these are things you've thought about:
- When prompts get really long or potentially contain long content, how to prioritize the contents to fit within the context
- How you might create an integrated evaluation flow within Rivet (there are so many edge case errors in AI agent dev and you'd want to cover as many as you can!)
Looks like a cool product! Will have to try it out on my AI projects.
@walden_yan1 thank you! And great questions!
We're far from done with Rivet! As we continue to work on Ironclad Contract AI, we will continue to add to Rivet's capabilities. AND we JUST added the plug-in system, which our friends at AssemblyAI have already been able to use to integrate their audio transcription and understanding AI models into Rivet! Tons of potential there.
Re: long prompts - Rivet doesn't do much to protect you from this automatically. However, for Ironclad Contract AI, we make use of things like the "chunk" node to split up large bits of text, and the Vector DB nodes for recalling relevant context. So we think there's enough to get by here, but there is certainly room for more.
Re: integrated evaluation flow - Rivet has "testing" utilities, but they're much more like "unit tests" than evals. While I could see this becoming "native" to Rivet, I actually think the plug-in system opens up the possibility of integrating Rivet with emerging solutions around evals!
Kudos on launching Rivet, it sounds like a groundbreaking tool in the LLM and AI agent ecosystem! Being open-source is such a power move - it truly lets a community sculpt the best version of a product. Given that we often deal with video production for SaaS companies, I'm eager to explore if there's a space where Rivet could revolutionize the way we interact with LLMs in this domain. Keep up the awesome work, and can't wait to see how Rivet evolves with community insights! 🌟
Could you share more about the specific use cases or industries where Rivet has proven to be particularly valuable? Additionally, what are some of the key features that set Rivet apart from other visual programming environments for AI agents?
This kind of innovation is essential as AI integration becomes increasingly important in various applications and industries.
@ricardo_luz great questions!
We've been working with other SaaS companies like Attentive, Sourcegraph, AssemblyAI, Willow, and Bento, who have tried using it in a variety of contexts. The main thing in common is that these teams all have substantial existing applications and codebases.
What sets it apart from other visual programming environments? Rivet is built to integrate with existing applications. More specifically that means:
- Remote Debugger. You can connect the Rivet UI to an app running locally or on a staging instance, and actually have it execute in that environment. Super great for iterating within your application, and you can even "hot reload" your agent graph.
- External Calls. You can dynamically define functions that your agent can call in the context of your application. This is especially great if you have access control concerns, because you can dynamically define the calls your agent can make, rather than give them unfettered access to an API.
- Collaboration via YAML Project Files. Rivet saves agent graphs to a YAML format. At Ironclad, we version these agent graphs with the rest of our git repo. We now regularly do code reviews on changes to prompts or agent graphs!
We literally could not have launched Ironclad Contract AI (https://ironcladapp.com/product/...) without Rivet.
Thanks for your support!
Super cool. What features do you think companies should build with agents using Rivet?
@doug_safreno1 thanks! At Ironclad, we have been using Rivet to build an agent in a chat interface. But we've already seen other companies doing very different things. Our friends at Bento (also on PH today!) created an awesome way to build new contextual on-boarding guides based on things like help center articles and just clicking around in an app.
Rivet