Today we're launching LGND Studio — an agentic workflow product that turns a plain-language request into a labeled, production-ready dataset.

Tell Studio what you need to find:

• "Find solar farms across Texas."
• "Show me signs of recent flood damage along the Gulf Coast."
• "Find more locations that look like this airstrip."

Studio takes it from there: searches imagery, surfaces likely matches, learns from your feedback, applies that learning across your full area of interest, and exports the results with provenance attached. Work through the Studio interface, or give the same capabilities to your own agents through our MCP server. Either way, you describe the outcome and get back data ready to use.

From a Question to a Dataset

Remote sensing is powerful, but the path from "I want to understand X" to "I have a labeled dataset" is normally long — juggling APIs, writing code to filter imagery, hand-labeling thousands of tiles, hoping the labels generalize. It can take weeks or months before you get an answer.

Studio closes that gap. You bring the question; the agent handles the orchestration — search, ranking, model training — and hands back the answer, the data, and a repeatable project you can refine or run again.

Better Data, Faster

Better. You train the model interactively and see exactly how it scores each candidate, so accuracy comes from your judgment, not a black box. A few positive and negative labels teach Studio the distinction you care about, not just the words in your original request.

Faster. Building a dataset drops from months to hours or less — no mapping expertise, no tile-by-tile labeling, no invisible batch processes.

How It Works

Studio runs on a single durable loop — seed → review + rerank → apply → export:

1. Seed. Describe your target in a sentence, drop in an example image, or point at an area of interest.
2. Review + Rerank. Mark a few thumbs-up / thumbs-down — no large labeled set required to start. The agent retrains automatically on your feedback and re-scores everything in real time.
3. Apply. To create a complete dataset across your full region, the agent runs the learned model — searching by meaning, resolving plain-English places and times ("California, last summer"), and handling multiple areas, time windows, and collections at once.
4. Export. Get a finished file with full provenance: what you labeled, what the model scored, and where every result came from.

Check out an example of creating a construction focused dataset in Huntsville, Alabama.

For Analysts and Developers

Analysts and teams get the chat-first loop shown above in a SaaS web application with durable projects and enterprise authentication.

Developers and their own agents get the same capabilities — semantic search, labeling, model training, apply-to-corpus — through an MCP server, so any LLM that speaks MCP can drive Studio directly:

• Embed Studio in your AI stack. Give Claude, or any MCP-capable agent, Studio's search and labeling tools as part of its own reasoning loop.
• Automate dataset pipelines. Workflow can be saved and run programmatically.
• Feed downstream systems. Pipe Studio's output straight into your analytics or ML pipelines.
• No default interface. All actions can be delegated by an agent, allowing for easier integrations into your ecosystem.

Same agent, different doors in — one for people, one for other agents.

Get Access Now

Studio is available now for Enterprise customers, contact Sales at sales@lgnd.ai for access.

What's Next

We're shipping updates weekly this quarter, keep an eye out for these upcoming releases:

• A more capable map — the agent works right on the map with you: reading what's on screen, labeling it directly, switching views, and dropping pins on the places that matter.
• Automated QA — the agent reviews your results as they come in, flagging edge cases and clearing out obvious false positives before you export.
• Segmentation — group tiles into real-world objects, turning chip boundaries into clean feature outlines.
• Imagery selection — the agent picks the best imagery for each question: clearest view, right season, highest resolution.