Agents Videoagents/video
Agents Videoagents.org.in / video·Private preview

Make video
like you write code.

Type a prompt — Sonnet 4.6 writes the composition. Hit render — WebCodecs and ffmpeg.wasm encode it on your machine. We never touch your media.

For Claude Code & Cursor usersFor creators who never open an editor0 frames sent to our servers
hook-v3.mp4· 1.8 MB · 58s render
product-hook.mp4
1920 × 1080·30 fps·ready

01 · the hook

Stop re-recording.

Ship the demo on the first take.

02 · proof

Browser-only render.

30 seconds · 1080p · 60s on a 2024 laptop.

encoding
92%
78%
64%
88%
webcodecs · 60 fps encode
●agents · video
typeseeship
agents · video
00:00
00:09

A 30-second hook for an AI screen recorder. Drafted by Sonnet 4.6 from one sentence, tweaked twice in chat, rendered locally in 58 seconds. No timeline-scrubber black magic required.

How it works

From a sentence
to an MP4 on disk.

01 · Describe

A sentence is the IDE.

“30-second product hook with bouncy captions and a voiceover.” Sonnet 4.6 reads it, drops a draft into the editor with timing, captions, and asset slots wired. The composition is plain HTML — readable, diff-able, version-controlled.

/new

30s product hook for an AI screen recorder, bouncy captions, female calm voice

composition.htmlpreview live
<Scene t="0-3">
<h1 class="hook bouncy">
Stop re-recording.
</h1>
<Caption style="marker" />
</Scene>
<Scene t="3-30">
<Voiceover voice="bella" />

02 · Refine

Edit it like you talk.

“Make the title 2× bigger.” “Swap the female calm voice for something gravelly.” “Add a 3-second outro card.” The agent writes the diff, you see the preview update in under two seconds, and you keep going. No timeline, no nested clip menus.

Make the title 2× bigger and swap the voice for something gravelly
−font-size: 56px;voice="af_bella"
+font-size: 112px;voice="am_michael"
Updated. Preview refreshed in 1.4s.
Add a 3-second outro with the URL
Scene 3 added · 28→31s total

03 · Run

TTS, captions, bg-removal — without a server.

Kokoro generates the voiceover. Whisper transcribes it into frame-accurate captions. u2net removes the background from a portrait clip. All three run via ONNX in your tab — cached on first use, instant after. No upload, no GPU bill, no waiting in a render queue.

Kokorovoiceover

ready · cached 325 MB

Whispertranscribe

9.4s clip → 12 captions

u2netbg removal

portrait.mp4 · 5.1s

0 bytes uploadedONNX Runtime

04 · Ship

Render in your browser. Pinned forever.

WebCodecs and ffmpeg.wasm encode the MP4 on your machine — 30s at 1080p clears in under 90 seconds on a 2024 laptop. Each project pins its runtime version. The video you ship today still renders pixel-for-pixel in 2030.

Rendering

s · 30s clip

codec

h.264

encoder

webcodecs

container

ffmpeg.wasm

Runtime pin

v3.4.0 · locked
v3.0v3.2v3.4v3.6v4.0

The render engine ships beside the runtime. Two projects on different versions render side-by-side in the same tab.

What ships in v1

Everything in the box.

Marketplace · seeded with 30 templates

Fork a template,
ship in five minutes.

v1 ships with 30 templates pulled from the public registry — product hooks, code walkthroughs, social cards, lower thirds, caption packs, and full scenes. Each one pinned to its runtime version, so nothing rots. Creator payouts open in v1.x.

Social

Product hook

Dev

Code walkthrough

Broadcast

Lower thirds

Captions

Caption pack

B-roll

Founder note

Endcards

Outro card

  • 01

    Workspaces

    Clerk Organizations

    Bring teammates into a shared workspace. Shared billing, member roles, and (soon) shared template libraries.

  • 02

    Multi-format export

    mp4 · webm · gif · png seq

    Same composition, four shipping formats. PNG sequences for downstream editors. WebM for the open web. GIF when you have to.

  • 03

    BYOK

    Anthropic, encrypted

    Drop your own Anthropic key in settings. Chat tokens never count against the built-in monthly cap. Encrypted at write time, never logged.

  • 04

    MCP server

    v1.1 · summer 2026

    Connect Claude Code or any MCP-aware agent directly to the same data model. Drive new projects, edits, and renders from your editor.

Web platform, not magic

Every preprocessing step runs on the device the user already paid for. We didn't invent the render path — we wired the pieces together and pointed an agent at it.

WebCodecsffmpeg.wasmONNX RuntimeKokoroWhisperu2netOPFSWebGPUClaude Sonnet 4.6MCPv1.1Clerk Billing

Pricing

Free is the floor.
Pay for the parts we host.

Your browser does the work. We bill for chat tokens, marketplace payouts, and the longer renders that need bigger ML models. Cancel from the account drawer — no clawback email, no calendar booking, no "please don't go" survey.

See the full table

Free

$0forever

Local renders, browser preprocessing, BYOK chat.

  • 5 clean renders / day
  • Or unlimited with a small watermark
  • 1080p · up to 30s

Hobby

$19/ month

For the indie shipping a video a week.

  • 200 renders / month
  • 1080p · up to 5-min clips
  • Built-in Sonnet 4.6 chat

Pro

$49/ month

For studios and devs shipping into production.

  • Unlimited renders
  • 4K · up to 30-min clips
  • Marketplace creator payouts
  • MCP access (v1.1)

From the founder

I wanted a Cursor for video — type, see, ship. Every tool I tried either ate my media, locked me into a timeline, or charged per render. So I'm building the one I'd actually use, on top of an open runtime that'll outlive any one company.
Harshit KhemaniFounder · AgentsORG
Agents Video

Private preview · summer 2026

Get in line.

Drop your email — one short note when your spot opens. No drip campaign, no newsletter, no "just one more thing".

We never sell your email · powered by Clerk