Read every layer
Text, captions, images, linked pages, speech, and video frames become a transparent coverage manifest.
Paste one public link. CivicLens detects every language automatically, reads captions and media, and shows the evidence behind each assessment.
Every checked public link has a shareable room for evidence, discussion, camera, and browser-tab video/audio.
CivicLens does not ask a model to guess what is true. It separates extraction, multilingual retrieval, and evidence assessment into visible steps.
Text, captions, images, linked pages, speech, and video frames become a transparent coverage manifest.
Queries run in each claim’s own language and English, with official sources matched to jurisdiction.
Verdicts require cited evidence. Missing evidence becomes a limitation—not an invented answer.
Every public link maps to one persistent room. YouTube supports host playback controls; articles and social links can be opened in a tab and shared with video/audio.
Confidence is not presented as a truth probability. CivicLens reports evidence strength, primary-source availability, and why each source was selected.
Every transcript segment and claim is detected independently. Original wording stays visible; evidence retrieval expands into local official languages and English.