Understanding what a satellite can tell us often begins with a search through papers, instrument documentation and data-product descriptions. ESA's Earth Virtual Expert, or EVE, puts a question-and-answer interface over that knowledge. The agency released the system on 8 October for researchers, policymakers, journalists and other interested users.[1]
EVE was developed by ESA Φ-lab with Pi School, Mistral and Imperative Space. Its current foundation is text: material from ESA, NASA and Copernicus, technical sources and research literature, including a collaboration with Wiley. Users can ask about topics such as vegetation change or how satellites monitor greenhouse gases.[1]
An answer connected to a specialist source collection
The system uses retrieval-augmented generation. Alongside what the model learned during adaptation, it can search additional material and bring relevant passages into the answering process. That provides a route to more recent evidence and domain-specific details that may be difficult to recover from a general model's training alone.[1]
Retrieval is particularly useful in Earth observation because the details of an instrument, processing method or measurement period can change the answer. A map of vegetation, for example, needs to be interpreted with its spatial resolution, time coverage and measurement assumptions in view. A specialist guide can help a reader discover which of those questions to ask.[1]
ESA also describes an LLM-as-a-judge stage that reviews generated responses and can flag material for revision or filtering. This is a reliability mechanism within the system, rather than independent confirmation that every answer is correct. The cited scientific source remains the place to inspect a consequential technical claim.[1]
What is available and what is planned
EVE's release and development path
| Stage | Capability or artifact |
|---|---|
| Live release | Text-based questions and answers about Earth observation |
| Public research materials | Model, code, curated datasets and benchmarks on project repositories |
| Second phase, begun September 2026 | Development of data retrieval, analysis and multistep tasks |
| Planned for 2027 | Searching and presenting satellite imagery |
The public model, code and supporting research materials make the project more inspectable than a closed demonstration. Developers and researchers can examine the available artifacts and compare future releases. ESA says regular updates are planned, so the distinction between an announced feature and a released one will remain important.[1][2]
A guide to the analysis before the analysis itself
A useful current interaction might ask which satellite measurements are relevant to a vegetation question, what a particular index represents or why two published estimates differ. Those are literature and interpretation tasks. Retrieving imagery, applying cloud masks, calculating a change and checking the result is a larger workflow that the second phase aims to support.[1]
That separation gives EVE a clear role now. It can help people approach a technical field whose information is scattered across many kinds of documents. The value of the interface is not only a shorter answer; it is a better route into the methods and evidence behind the answer.[1]
As the project adds tools for data and imagery, the connection between explanation and computation will become more direct. The present release establishes the language and retrieval layer through which users can begin that work, while its open artifacts give the research community a way to examine how the guide is built.[1][2]
Sources & further reading
- Introducing ESA's Earth Virtual Expert European Space Agency · 8 Oct 2026 · Accessed 27 Oct 2026
- Earth Virtual Expert — open-source project repositories ESA Earth Virtual Expert GitHub · Accessed 27 Oct 2026