02 · THE QUESTION
Ask a question. DEVLOK finds the evidence.
“Has the built-up area increased, and where?”
03 · THE AGENT
The agent understands, then selects.
QUERY↓ TASK CLASSIFICATION↓ INPUT VALIDATION↓ MODEL / TOOL SELECTION
04 · OPTICAL + SAR
Two sensors. One understanding.
Optical gives spectral, contextual information. SAR adds structural detail — through cloud, day or night.
OPTICAL+ SAR→ COMPLEMENTARY INFORMATION→ FUSED UNDERSTANDING
05 · TIME LEAVES EVIDENCE
The same ground, re-observed.
2018→2020→2022→2024→2026
WHAT changed? WHERE? WHEN? Bi-temporal imagery reveals it.
06 · EVIDENCE VERIFIED
Change, located and grounded.
▲ CHANGE DETECTED
Scroll on for the verified answer, the highlighted region, and the execution trace.
DEVLOKORBITAL DESCENT AOI12.97°N 77.59°E ALT680 KM (LEO)
DEVLOK · AGENTIC EARTH-OBSERVATION INTELLIGENCE

SEE EARTH
DIFFERENTLY.

Ask a question. DEVLOK finds the evidence.

QUERY“Has the built-up area increased, and where?”
RESOLUTION0.30 M/PX
STAGE01 · SEE EARTH
SCROLL TO DESCEND
BUFFERING FRAMES: 0% (0/240)
MISSION

Satellite imagery holds answers. DEVLOK turns a natural-language question about Earth into an evidence-backed remote-sensing answer.

USER→QUESTION→DEVLOK→EVIDENCE

Start with plain language.

No GIS workflow to assemble, no model to hand-pick. Ask about what you see — DEVLOK decides which data and models the question needs.

NATURAL-LANGUAGE QUERY
“Has the built-up area increased, and where?”

One interface for single images, optical–SAR pairs, and bi-temporal observations. The query carries the intent; the agent carries the plan.

THREE INPUT MODES
MODE A
Single image
→
MODE B
Optical + SAR
→
MODE C
Bi-temporal

Every answer returns text, spatial evidence, and an execution summary.

The agent decides.

Adapted from satquery-agent-decides.html (SatQuery project reference) — the query determines the workflow, not the other way round.

↓
↓
↓
↓
↓
↓

One image, three skills.

The specialist bench for single optical, multispectral, or SAR frames.

◉
Visual Question Answering

Ask pointed questions about land cover, objects, and layout in a single observation.

VQA SPECIALIST
☰
Captioning / Scene Description

A grounded natural-language summary of what the sensor actually recorded.

DESCRIBE
⌖
Text-guided Grounding

Link phrases to pixels — every claim points at a region you can inspect.

GROUND

One image isn’t always enough.

Adapted from satquery-crossmodal.html (SatQuery project reference). Drag the fader — the same ground read twice, then understood jointly.

Optical satellite viewOPTICAL
SAR rendering of the same sceneSAR
◀ OPTICAL · SPECTRALSTRUCTURAL · SAR ▶

VISUAL DEMONSTRATION — illustrative fusion preview, not a live model run.

JOINT UNDERSTANDING

Complementary, then combined.

OPTICAL gives spectral, contextual information — land cover, vegetation, texture. SAR provides complementary structural information and keeps working through cloud cover, day or night. DEVLOK aligns both to shared coordinates, then reasons over the fused layer.

OPTICAL+SAR→COMPLEMENTARY INFORMATION→FUSED UNDERSTANDING

Time leaves evidence.

Adapted from time-leaves-evidence.html and satquery-find-the-change.html (SatQuery project references). The same geographic scene, 2018 → 2026 — watch buildings, roads, and land use change, then see the change mask. Drag the divider; tap a highlighted region.

Earlier observation (T1)
Later observation (T2)
↔
T2 · LATER T1 · EARLIER
20182020202220242026

WHAT CHANGED? · WHERE? · WHEN? — the change mask answers all three.

06 · FIND THE CHANGE

Something changed here.

Three candidate regions survived the bi-temporal pass. Select each one to reveal its evidence.

Change-analysis frame with candidate regions
REGION A · CHANGE DETECTED
REGION B · NO SIGNIFICANT CHANGE
REGION C · WATER EXTENT CHANGED

Select a region above.

VISUAL DEMONSTRATION — illustrative imagery and regions.

Yes — and here is where.

The verified result for “Has the built-up area increased, and where?” No invented confidence or benchmark scores — only the trace.

DEVLOK // VERIFIED RESULT08 · EXECUTION TRACE ↓
QUESTION
“Has the built-up area increased, and where?”
ANSWER
YES — BUILT-UP AREA INCREASED · REGION A
TIME INTERVAL 2018 → 2026 observation window
TASK      Change-based VQA
MODELS/TOOLS Change Understanding · Grounding · Optical–SAR Analysis
INPUT       Bi-temporal, co-registered frames
OUTPUT      Answer + change mask + grounded regions + this trace
CONFIDENCE  Produced at inference time — not pre-filled here.
Evidence frame with highlighted change region ▲ CHANGE DETECTED · REGION A

Tuned for orbit, not for the web.

Generic vision-language models meet remote-sensing data — then serve inside the DEVLOK agent. Datasets below are training and evaluation resources, not performance claims.

STEP 1
Generic VLM
→
STEP 2 · ADAPTATION DATA
BigEarthNet
→
STEP 3
RS-adapted component
→
STEP 4
DEVLOK agent
EVALUATION RESOURCES
VRSBench
Vision-language evaluation for remote sensing.
RSVQA
Visual question answering benchmark.
CDVQA
Change-based VQA benchmark.
ISRO / SAC set
Co-registered Cartosat-2S optical + RISAT SAR evaluation pairs.

One Earth. Many questions.

From satquery-mission-selector.html (SatQuery project reference). Pick a mission — its question loads straight into the analysis terminal.

Ask DEVLOK.

A preview of the DEVLOK console. Chips fill the prompt; Run walks the demo trace.

devlok — analysis console ● READY
devlok>
TRY:
TRACE — press Run Inference to walk the DEVLOK pipeline.

PREVIEW CONSOLE — the trace below is simulated for demonstration.

01
Task routed
CHANGE-BASED VQA
03
Specialists
CHANGE · VQA · GROUND
02
Frames compared
T1 → T2
∞
Zoomable evidence
MASKS + BOXES

Making Earth-observation intelligence accessible through natural language.

Orbit, ask, verify. The demo pages stay where they are — this page is the mission.

Open agent demo →
RAVIMission Guide
Ask a question about the Earth.