Underqubit Blog
AI, software engineering, and the technology behind Underqubit's services — for organizations where people and AI agents work together.
EngineeringSidecar MCP: turn your REST APIs and CLIs into MCP tools with a YAML file
We're open-sourcing Sidecar MCP under Apache 2.0: an MCP server that exposes existing REST APIs and CLIs as tools declared in YAML or JSON, with no code.
Read more
AISystem One Models: What They Are and When to Reach for Jev Instead of an LLM
Jev, TypeSafe AI's System One model, doesn't generate text — it returns typed, calibrated decisions in milliseconds. What it is, when to use it, and best practices.
Read more
CybersecurityTop 10 AI Agent Attack Vectors (and How to Mitigate Them)
AI agents don't just hallucinate — they can be hijacked and used as a pivot into critical systems. The 10 attack vectors that matter most, and how to mitigate them.
Read moreEngineeringSpec-Driven Development: Best Practices, Techniques, and When to Use Each
AI agents write code fast and interpret intent poorly. Spec-driven development fixes the second problem — here's how to do it well, and which of its three techniques to reach for.
Read moreAIHow to Test a RAG System: Metrics, Datasets, and Tests That Catch Real Failures
A RAG system is not validated because its answers sound good. Use these metrics, tests, and traces to measure it before and after production.
Read moreAIMultilingual RAG: Techniques and Best Practices for Retrieval Across Languages
When your users and your documents don't share a language, retrieval-augmented generation gets a lot harder. Here's what actually works.
Read more