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

# Introduction

> Lavendly is an AI video platform where an agent produces a finished film (script, voiceover, music, captions, and cut) without a human in the loop.

Most AI video tools generate a clip. **Lavendly composes a film.**

The product is a graph: nodes are intents (text, images, voices, music,
shots), edges are how they compose. The same graph that a human edits on
the canvas is the one an agent edits through the API, and either side
sees the other's changes in real time.

<CardGroup cols={2}>
  <Card title="Quickstart" icon="rocket" href="/quickstart">
    Get an API key, generate your first video in five minutes.
  </Card>

  <Card title="Agent skills" icon="robot" href="/agent-skills/overview">
    Drop Lavendly into Claude, Cursor, or any agent runtime.
  </Card>

  <Card title="SDKs" icon="code" href="/sdks/node">
    Node and Python cookbooks, generate, narrate, render, ship.
  </Card>

  <Card title="API reference" icon="book" href="/api-reference/overview">
    Every operation, idempotent where it matters.
  </Card>
</CardGroup>

## What "agentic-first" means

We built the API first, then put a canvas on top. Both edit the same graph:

* **Same surface, same trust.** Every action a human can take, an agent
  can take. There is no "agent backdoor" and no "human-only feature."
* **One source of truth.** When an agent attaches a voiceover or changes
  the mix, it shows up as a real node on the canvas, instantly editable
  by a human, deletable by either side, rendered the same way.
* **Idempotent by default.** Retry-safe headers on every mutation that
  costs money. Agents can crash, networks can blip, and the same call
  twice still spends once.
* **Self-describing.** A single endpoint returns the complete operation
  catalog so an agent can discover what the platform can do without
  reading docs out of band.

## What an agent can build with it

<AccordionGroup>
  <Accordion title="A narrated short from a one-line prompt" icon="film">
    Generate a shot card, pick a video model, synthesize a voiceover with
    word-level timing, choose a music bed with ducking, mix it all,
    render. \~6 tool calls end to end.
  </Accordion>

  <Accordion title="A multi-clip story with consistent characters" icon="people-group">
    Spin up several shot nodes, share a character anchor across them,
    chain them into an output node, render. Characters, palette, and
    motion vocabulary carry across the sequence.
  </Accordion>

  <Accordion title="A localized variant pipeline" icon="globe">
    Take a finished workflow, swap the voiceover for a different
    language and voice, re-render. Original graph untouched.
  </Accordion>

  <Accordion title="A weekly auto-publish loop" icon="calendar">
    A cron'd agent reads a feed, drafts a workflow, renders, posts the
    finished video. Credit ledger and monthly cost cap keep it from
    bankrupting you on a bad week.
  </Accordion>
</AccordionGroup>

## Where to start

* **Building an agent?** [Agent skills](/agent-skills/overview), then the [Node](/sdks/node) or [Python](/sdks/python) SDK.
* **Integrating into your product?** [Quickstart](/quickstart) and the [API reference](/api-reference/overview).
* **Just evaluating?** Run the [first-video guide](/guides/build-your-first-video) end to end.