[ Journal ]
Notes from the prompt frontier.
Engineering deep dives, research and product updates from the team building PROMPT ZERO.
Latest posts
Research
Routing across 40 models without losing your mind
How our router picks the cheapest model that still passes your evals, and what we learned from 4 billion routed tokens.
Product
Agent Runtime 2.0 is here
Durable tool calls, resumable runs and step-level traces. Everything you need to put agents in front of real customers.
Guides
Designing golden datasets that actually catch regressions
A practical guide to building eval sets that reflect real traffic instead of the demo you gave last quarter.
Research
LLM judges are biased. Here is how we calibrate them
Position bias, verbosity bias and self-preference, measured across 12 judge models, plus the fixes that worked.
Guides
Self-hosting PROMPT ZERO in an air-gapped VPC
Step by step: from Terraform to your first traced request, with zero outbound network calls.