> ## Documentation Index
> Fetch the complete documentation index at: https://docs.vidjutsu.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Features

> Everything your agent gets with VidJutsu.

## Watch — AI video understanding

Send a video and a freeform prompt. AI watches the full video and answers whatever you ask — hook quality, pacing issues, content summary, brand safety. No rigid schema, just a question and an answer.

## Extract — server-side processing

Pull frames, audio tracks, and metadata from any video. Server-side processing — no local install required. Auto-select keyframes or request specific frame counts. Get resolution, duration, FPS, and codec info.

## Transcribe — speech-to-text with timing

Full speech-to-text with word-level timestamps. Know exactly what was said and when. Use it for subtitle generation, content verification, or feeding transcripts into downstream analysis.

## Check — spec validation

Validate specs against built-in rules and your own custom rules. Define rules like `hook-required`, `max-duration-60s`, or `aspect-ratio-9-16`. Save per-client custom rules via the API and run checks programmatically.

## Compliance — platform TOS pre-flight

Scan any video or caption against the target platform's current policies before you publish. Covers **YouTube, TikTok (consumer + ads), Instagram / Meta, Facebook Ads**. Returns a deterministic 0–100 risk score, a verdict, and each violation with the **verbatim policy clause** and source URL it's based on.

* Anti-hallucination validator — every cited rule is resolved against our DB before returning; the LLM can't invent clauses
* Two inputs: `POST /v1/compliance/video` (scans a hosted video URL, 20/day) and `POST /v1/compliance/prompt` (scans caption or ad copy, 100/day)
* Every response includes a `policySnapshotDate` timestamp and a disclaimer field
* Informational only — not legal advice, and no freshness or accuracy guarantee. The citation reflects what we last ingested from the platform's own published text; you're expected to verify the live source before acting on the result. See [Terms of Service](/legal/terms)

## Storage — CDN and tagging

Upload videos to a global CDN with `POST /v1/upload`. Organize your content library with accounts, posts, and AWS-style key-value tags. Filter anything by tag — `?tag.brand=acme`, `?tag.batch=week-12`.

## Composable

Use the full stack or just parts. Watch a video without storing anything. Extract frames without transcribing. Every endpoint works on its own — no required workflow, no orchestration layer. Build the pipeline that fits your agent.