Product
Cascaid
Predictive cascading-failure intelligence for agentic and RAG pipelines.
§1 — The problem
Tracing tools show you what already broke. Cascaid predicts what is about to.
Observability tools like Langfuse, LangSmith, Arize Phoenix, and W&B Weave are diagnostic: they trace a pipeline after a failure so you can understand it. Cascaid is predictive: it watches the graph of an agentic or RAG pipeline in production and flags the component that is about to take the rest of the system down with it, before it happens.
§2 — How it works
Graph Store and a GNN model, trained on your pipeline
Instrument
Auto-instruments LangGraph, LiteLLM, Pinecone, and Weaviate, no SDK or decorators.
Graph Store
Pipeline telemetry is modeled as a graph of components and their dependencies.
GNN Model
A graph neural network trained on that telemetry predicts which node is about to fail.
§3 — Quickstart
Self-hosted. Your data never leaves your environment.
Apache 2.0 licensed. Run a zero-setup synthetic demo, or bring up the full dashboard, model server, and Postgres against a real pipeline.
pipx install cascaid
cascaid demo