Create infinite experiences, in minutes.

Worldstreams in action

Rick and Morty have their own character controls, and both actions feed the shared experience. Watch the multiplayer Worldstream.

What GENESIS handles

You define the experience. GENESIS runs the state and interaction loop around your connected models.

• World memoryRemember choices, quest progress and relationships from one turn to the next.
• Inventory and variablesTrack items, health, scores and resources using your world's rules.
• Characters and locationsKeep your cast, places and scene facts organized for the next generation request.
• LLM prompts and repliesPass world context to your LLM and validate structured moves, dialogue and choices.
• Custom interfacesBuild player controls, inventory panels and overlays with HTML, CSS and JavaScript.
• Multiplayer and shared rulesCoordinate player seats, turns and crowd votes. Keep everyone's accepted state in sync.
• Scene continuityTrack generation jobs, carry the actual last frame forward and advance after playback.
• Portable worlds and assetsExport your settings, interface, controller and declared media in one complete bundle.

Quickstart with the Genesis skill

Give your coding agent an idea. The included create-world-stream skill guides it through building and checking a playable prototype.

1. Open Genesis in your coding agent

   Clone the repository and open its folder in your agent. The skill is included at skills/create-world-stream/SKILL.md. View setup commands.

2. Describe your world

   Start with the prompt below. Change the setting, controls and visual style to make it yours.

   Read skills/create-world-stream/SKILL.md and use it to build Mosslight, a forest battle worldstream with a cream pixel interface.
   
   Let two players take turns. Have the backend LLM create four moves per encounter. Remember inventory, health and progress.
   
   Start with labeled local fixtures. Validate the project, test the controls, and give me a playable local preview and a complete ZIP export.

   Copy world-building prompt

3. Play, refine and export

   Open the local preview, try the controls and ask your agent for changes. Your ZIP contains the world's code and declared assets, ready to import into a compatible host.

Local previews use prepared responses and media. For live generation, connect the Worldstreams API or your own models.

Follow the full skill quickstart · Run a starter yourself

What can you build?

One engine, many kinds of infinite experience. Start with a world people can play, watch, explore or learn from.

Play

Games, multiplayer games and creative sandboxes. Build quests, cooperative battles and worlds players can keep changing.

Watch

Infinite TV, interactive stories and live events. Create ongoing shows and performances shaped by the audience.

Explore

Travel experiences, social worlds and virtual showrooms. Visit destinations, meet characters and tour spaces together.

Learn

Learning experiences, interactive history lessons and robotics training simulations. Practice languages, step into another era or connect specialist simulation tools.

Practice

Coaching, training and rehearsal, and therapy and wellbeing. Rehearse interviews and workplace scenarios, or build practitioner-led reflection experiences.

You supply the domain content, rules and any specialist tools your experience needs.

Three files. Your whole world.

my-world/
├── genesis.json      World settings, actions, tools and assets
├── interface.html    Your interface and player controls
├── controller.js     World memory, rules and backend requests
└── assets/           Images, audio and video that travel with you

GENESIS coordinates the world; connected providers generate media. Your host supplies identity, storage, credentials and spending policy. Start building.

Choose your connection

To run a live world, you need hosting, persistent state, user access, billing controls, an LLM and a World Model. Choose which parts you want to operate yourself.

Scroll the table horizontally to see every option.

Choose BYO for full control over your stack, Sequencer API for managed generation inside your app, or Sequencer hosted to create and publish using an existing account and workspace.

LLM and media services are available through the same Sequencer key. Custom LLM calls, images and audio have their own usage charges. Your own app remains responsible for its audience's authentication, database and monetization.

Bring your own providers · Connect the Sequencer API · Create or import in My Worlds

Compare Worldstream models

Compare native video, published pricing and how quickly a model can respond. The hourly figures use the provider's billing unit, so they show model cost rather than the total cost of running your app.

Scroll the table horizontally to compare all model details.

Rates checked October 13, 2026. Reactor bills time holding a GPU, including idle time. fal Director bills generated video with a 60-second minimum. Sequencer advertises active stream time; use its session quote for the current rate. LLM tools, extra assets and your app infrastructure can add costs.

Quality ratings use a shared five-point rubric. Models remain unrated until they have been tested with the same prompts and conditions. See the sources, billing details and rating method.

The source is Apache 2.0 licensed. Version 0.1 is a source preview, with npm publication and the wider launch being prepared. Browse the repository, read the license, contribute or explore the SDK API.