Ironclad/rivet

What it solves

Rivet simplifies the creation of complex AI agents and prompt chaining by providing a visual IDE. It removes the need to manually code intricate LLM workflows, allowing developers to design, test, and then embed these AI behaviors directly into their own applications.

How it works

It consists of two primary components:

1. Rivet Application: A desktop IDE where users can visually build AI agents and prompt chains as graphs.
2. Rivet Core: A TypeScript library (@ironclad/rivet-core) that executes the graphs created in the IDE. This library allows the AI logic to be integrated into external applications, enabling bidirectional communication where the application can trigger Rivet graphs and the graphs can call back into the application's code.

Who it’s for

Developers who want to build sophisticated AI agents and prompt-based workflows without writing all the orchestration logic by hand, and who need a way to deploy those workflows into a production application.

Highlights

• Broad LLM Support: Integrates with OpenAI (GPT-3.5, GPT-4), Anthropic (Claude family), and AssemblyAI (LeMUR).
• Vector Database Integration: Built-in support for OpenAI Embeddings and Pinecone.
• Multimodal Capabilities: Supports Speech-to-Text via AssemblyAI.
• Flexible Deployment: Graphs can be run via a TypeScript library for seamless application embedding.