Shippy is an AI-powered conversational assistant designed to help maritime analysts get more from the Skylight platform and its data ecosystem. Shippy can answer questions in natural language about vessels, maritime activities, and provide contextual guidance grounded in Skylight's capabilities. It is available within Skylight – currently only to select users.

What Shippy Does

Instead of navigating multiple buttons to manually retrieve complex insights from the Skylight platform, users can simply ask questions in plain language, such as:

• “Show me recent fishing events in the Nauru EEZ”
• “What do we know about this vessel’s movements over the past 30 days?”
• “Are there any dark rendezvous events near the Galápagos?”

Shippy translates these questions into structured queries against Skylight’s APIs, and returns actionable results, helping users quickly uncover the information they need. Shippy works in all languages.

How It Works

Unlike public-facing LLM tools, Shippy is purpose-built for maritime analysis and has direct access to the same real-time/near real-time data sources visible in Skylight such as AIS and satellite imagery. 

Think of Shippy as your maritime-specialized AI assistant: combining modern LLM capabilities with dedicated maritime expertise, domain-specific analytical reasoning, and access to real-time maritime data sources.

Specialized maritime skills

Shippy is designed to apply maritime concepts, terminology, and analytical methodologies to its reasoning. Its a pre-requisite for users to have a strong understanding of Skylight data to use Shippy effectively and correctly interpret its outputs.

Shippy generates insights using three data ecosystems, prioritized as follows:

1. Skylight Data Ecosystem (Primary Source)
   Shippy relies primarily on the Skylight data ecosystem, using its proprietary maritime datasets as the foundation for analytical outputs and responses.
2. Sanctioned External Data Sources
   For certain queries, Shippy may draw upon sanctioned external datasets from trusted, authoritative, and specialized sources to supplement Skylight data with targeted information relevant to the user's request.
3. Open Web-Based Information
   When appropriate, Shippy can leverage publicly available web-based information to enrich its responses and provide broader context surrounding a query.

Responsible by Design

The agent is designed to present factual, observation-based information rather than making legal determinations or characterizing vessel activity as inherently suspicious. In a world where generic LLMs often strive to provide an answer to every question, Shippy is aware of its own limitations and clearly communicates when it can not provide a reliable answer.

Shippy also systematically provides deep links so users can verify the information directly from the original source, whether it derives from the Skylight data ecosystem, sanctioned external datasets or open web-based information.

Data Security & Privacy

The Skylight program places a strong emphasis on protecting the security and integrity of its data, and Shippy is designed with the same data security considerations at its core. Shippy notably features:

• Account-encrypted access: Shippy is not a public tool and is only accessible to users and organizations that have been vetted by Ai2 / Skylight and granted authorized accounts.
• No cross-user visibility: each user accesses a distinct version of Shippy that can only retrieve data (Areas of Interest, watchlists, recurring queries) associated with that user’s account. Shippy won’t answer queries about what another user is doing with Skylight.

In its current version, your Shippy conversations are stored by Skylight and Anthropic but they are not publicly visible to other users. The Skylight team currently retains visibility on user’s queries to guide future product development, optimize performance, and adapt the system to user needs. It also enables traceability and review of any hallucinations or errors generated by Shippy.

Under Anthropic’s commercial API terms, inputs and outputs related to Shippy queries may also be retained by Anthropic for up to 30 days for trust and safety purposes, such as potential violations of their Acceptable Use Policy. However, under the current API terms, Anthropic does not use these inputs to train its models. These user conditions are broadly consistent with standard commercial practices across most LLM providers today and not something Skylight can derogate from.

Continuous Evaluation

To ensure Shippy meets the standards expected by Skylight's users and partners, the team runs a structured evaluation process. A suite of realistic test scenarios — covering vessel queries, fishing reports, guardrail compliance, and more — is executed against Shippy on a regular basis. Each scenario is scored by an LLM judge against criteria defined by subject matter experts, and the results are reviewed by the team to identify areas for improvement. This process helps us catch issues like over-attribution of suspicious intent, insufficient source crediting, or data exposure risks before they reach users.

How to use Shippy Responsibly

Like any large language model, Shippy can produce hallucinations. To avoid being misled, users should always evaluate its outputs critically and verify key findings against other sources of information or their own knowledge of their operational context.

A useful way to think about Shippy is as a junior analyst who has performed the initial data pulls: the underlying data may be correct, but the framing and interpretation still require senior-level review and judgment. Try to question its assumptions, validate its data as evidence, and ensure the conclusions it makes are fully supported by the data!

Getting Started

Shippy is being made available to Skylight users on a rolling basis. If you would like to try it out, reach out to your Skylight point of contact or support@skylight.global. We are actively refining Shippy's capabilities based on evaluation results and user feedback, with the goal of making maritime intelligence more accessible and actionable for everyone working to protect our ocean.