Semantic video search for Flutter, without the ML ops

What it does

VModal for Flutter is a Dart SDK that wires mobile apps to a multimodal video-search backend. It handles streamed uploads, progress tracking, cancellation, and natural-language search across visual moments, spoken dialogue, and on-screen text. Your app keeps the UI and credential storage; the SDK manages the gateway, request models, and scoped resource paths.

The interesting bit

The SDK deliberately owns zero interface. It never provides a login screen, file picker, or persistent key store, which is rarer than it sounds for a SaaS-backed SDK. Instead it exposes an immutable scope model—project, collection, stream—that maps cleanly to multi-user or catalog layouts, plus raw JSON passthrough so new server fields land immediately without waiting for a typed release.

Key highlights

• Search video by natural language across semantic image content, ASR transcripts, and OCR text
• Streamed, signed uploads with cancellable progress broadcasts and no full-file memory loading
• App-owned auth lifecycle: runtime API keys, rotation without client rebuild, and no imposed login UI
• Typed responses with stable contracts, while preserving raw JSON for forward-compatible field access
• Deterministic cleanup via project.close() and per-operation cancellation tokens

Caveats

• Mobile-only: Flutter Web, macOS, Windows, and Linux are explicitly out of scope for the 1.0 release
• Multipart upload is experimental and fails with FeatureDisabled if the backend route family is unavailable
• Requires a VModal API key and service backend; the SDK is a client, not a self-hosted engine

Verdict

Worth a look if you are building an Android or iOS Flutter app with searchable video libraries and would rather not operate your own ML indexing pipeline. Skip it if you need web or desktop support, or an entirely offline, self-hosted stack.