A simple canvas for language and audio models.

This is a place for quickly testing out different mixes of language and audio generation models. Everything is currently stored locally in your browser.

Preview

Connect third party services to start adding to blocks to your canvas. Your keys are kept in local storage. All calls to third parties happen between your browser and the third party service.

In the beta you won't need to enter any keys. We will provide these integrations for you. We will also provide a way to publish your workflows and share them with others.

The following example grabs the five most recent items from an RSS feed, creates a readable summary using mixtral-8x7b-instruct, and pipes the output to an ElevenLabs tts model.

Services

OpenRouter

ElevenLabs

Workflow

0
Use RSS Feed
Grabs the most recent {numItems} from an RSS feed {source}.
source
numItems
↓
1
Invoke LLM (OpenRouter)
Call language model {model} with {messages}. Use template `{previousBlockResult}` to insert the output of the previous block.
messages
You are a writer for the news. You create readable scripts for a news program. These scripts will be read aloud by one person. Do not include anything that won't be read aloud. Skip items that aren't newsworthy. Add a sign-off at the end of the script.
Create a script for me from the following: {previousBlockResult}
model
↓
2
Create Audio (ElevenLabs)
Generates an audio file from {text} using {model}. Use {previousBlockResult} to insert the output of the previous block.
textGood morning, here are the top stories from Techmeme. {previousBlockResult}
voice