Skylight Chat is a new feature that helps you ask questions about maritime activity using simple, everyday language. Instead of searching through maps and filters, you can type a question like:

“How many vessels were in the Ghana EEZ yesterday?”

…and Skylight Chat will find the answer for you.

This tool is built to make it easier for anyone—no matter your technical background—to explore vessel activity, spot trends, and get fast insights from Skylight’s data.

You don’t need to understand artificial intelligence (AI) or machine learning (ML) to use it. Just ask your question the way you would ask a person. The system will search the right data and return a clear, readable summary.

NOTE: This tool is in BETA. It may not always produce accurate answers. Please check any outputs before applying them to operational activities.

🔑 How to Use

1. Log into Skylight like normal
2. Go to: https://sc-production.skylight.earth/chat
3. Type your question
4. The chat will return a Query Plan to show you what information it got from your question and how it will answer it
5. Then the chat will return an Answer with additional details as a report

📦 What Types of Answers You Can Get

Skylight Chat gives four kinds of reports depending on your question. Each one shows a summary of vessel activity or detections. These are:

1. Area-Based Vessel Summary

Tells you how many vessels were in an area and what kinds they were.

You’ll see:

• Total number of vessels
• Vessel types (like fishing, cargo, or tanker)
• Vessel flags (their registered country)

Example Question:

“How many fishing vessels were in the Ghana EEZ yesterday?”

2. Area-Based Event Summary

Shows events that happened in a specific place during a certain time. 

You’ll see:

• Numbers of different events (like dark activity or rendezvous)
• Detections from satellite tools
• Vessel flags (which country the ships are from)
• Vessel types (fishing, cargo, etc.)

Example Question:

“Tell me about Dark Rendezvous Events in the Colombia EEZ yesterday."

3. Vessel-Based Summary

Gives information about a single vessel.

You’ll see:

• Name, MMSI, flag, type, and size
• Past activity (rendezvous, dark activity, detections)
• Extra information from partners like Global Fishing Watch, if available
• Event information, if relevant

Example Question:

“Summarize all fishing activity for MMSI 224081660.”

4. FAQ Answers

General information about the Skylight platform and data sources. 

Example Question:

“What satellites does Skylight have access to?”

💡 Why Use Skylight Chat?

• Faster: No more switching between tools—just ask your question.
• Easier: Use natural language. No special format required.
• Clear: Answers are short and readable, with links to more data.
• Powerful: Built on Skylight’s most accurate vessel and event data.

⚠️ What It Can’t Do (Yet)

Skylight Chat is still improving. Here are things it does not handle:

• ❌ General political questions (e.g., “What are the maritime laws in Indonesia?”)
• ❌ Speculative questions (e.g., “Is this vessel involved in smuggling?”)
• ❌ Questions with vague timing (e.g., “sometime last season”)

🧠 How It Works (Behind the Scenes)

Skylight Chat uses a system with several steps:

• Input Resolver – Understands what you’re asking.
• Planner – Decides what tools to use (like vessel search or event data).
• Tool Runner – Pulls the data from Skylight’s backend.
• Analysis Generator – Writes the summary based on results.
• Final Answer (Optional) – Adds a short answer when needed.

You don’t see these steps—but they all happen in the background to make your answer appear quickly and clearly.

📩 Sharing Feedback

Below each Query Plan and and Answer, there are feedback buttons to leave comments about how well the plan and answer did. Leaving feedback is critical to help us improve the tool and ensure it is producing accurate answers.

👍 If you like the plan or answer, give us a thumbs up.

👎 If Skylight Chat handles the question incorrectly, gives a wrong or inaccurate answer, takes too long, or misinterprets your question, give us a thumbs down. 

💬 In either case, but especially for 👎 , please click the comment button to tell us why you left this feedback.

⚙️ Technical Details

There are a lot of resources to learn more about Natural Language Interfaces like Skylight Chat and ChatGPT. All of these products are built off of Large Language Models (LLMs) which handle the conversion of human natural language into tasks and queries that computer systems can understand.

There are many LLM models today, including a fully open-sourced model developed by Ai2 (more here). The LLM that Skylight Chat uses is OpenAI's GPT-4o, which comes with a lot of supportive tooling like an Agents SDK which our development team found to be particularly beneficial to this implementation.