The architecture behind an agent whose wrong answer moves a patrol boat
- Source
- x.com
- Date
Our @SkylightMarine team is rolling out Shippy, an AI agent for the people protecting our ocean in real time. A wrong answer at these stakes can send a patrol vessel miles off course. Here's the architecture we built so analysts can trust it. 🧵
@SkylightMarine Shippy is an AI agent built on Skylight, our real-time maritime intelligence platform. Ask a question in plain language & it fuses data sources, cites them, and delivers clear maritime intelligence in minutes. Making it reliable came down to architecture.
@SkylightMarine Shippy queries Skylight through a purpose-built CLI with typed flags. It handles auth, pagination, & structured output, preventing dropped results + misread responses. Writing to disk – not piping through the shell – keeps large queries from breaking downstream tools.
@SkylightMarine Serving Shippy to hundreds of agencies across 70+ countries meant building new infrastructure. Mothership, our agent hosting platform, runs every conversation in its own ephemeral, isolated session, so no agency ever sees another's data.
Context
Reliable data retrieval is part of the design of Shippy, Ai2's maritime analysis . Early prototypes constructed their own requests to the underlying data service and sometimes mishandled filters or pagination, silently losing results. The team replaced that approach with a purpose-built command-line tool that handles access credentials, retrieves successive result pages and writes structured results to a file. Its engineering account explains how this makes data access predictable and separately testable, reducing the amount of request-handling logic the must improvise during an investigation.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
- eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
Checking sign-in…
Loading comments…