Every step is a node. Every wire is typed.
Drop a model on the canvas and drag its output into the next step's input. Text fans in, media stays exact, and the editor refuses anything the server would reject — so what you draw is what runs.
Run inputs, fixed values, wires
Ask the runner for a photo and a mood, bake the prompt that never changes, and let each step feed the next.
Watch it happen live
Every run streams step by step. Stop it, resume from the last good step, download every output.
A designer for captions
Place, style and animate captions on a real 9:16 preview, and choose which knobs the runner may turn.
Let heddla draft it
Describe what you want. heddla works the bench in turns, proposes a build, test-renders it, and hands you a working pipeline.
The best model for each job, one balance for all of them.
Gemini, GPT, Claude, Kling, Veo, Runway, ElevenLabs, Flux and a set of deterministic tools, on one balance with prices you see before you run. Usage-billed models are metered exactly as the provider bills them.
- Text and reasoning. Gemini, GPT and Claude for writing, extraction, translation and structured output.
- Images. Nano Banana and Flux for generation and edits, from one photo or many of the same subject.
- Video. Kling, Veo, Runway and MiniMax to animate a still or shoot from a prompt, plus trims, reframes and burn-ins.
- Voice and music. ElevenLabs and OpenAI voices, generated music, transcription, and audio laid under any clip.
- Tools. ffmpeg, ImageMagick, poppler and pandoc steps that run in seconds and cost cents.
Every model, its inputs and its price, on one page.
The catalog is public: what each model takes, what it returns, which options a pipeline can open to the person running it, and what a step costs before you commit to it.
Publish it, or call it.
Nominate a pipeline to Discover and anyone can run it on their own balance. Or mint a key and run it from your product — every pipeline has a docs page with the exact payload it takes. Agents connect over MCP and get the same operations as tools.
POST https://api.slai.se/v1/pipelines/{id}/executions
Authorization: Bearer slai_…
{
"inputs": {
"portrait": {"type": "image", "file_id": "…"},
"look": {"type": "text", "text": "Aged blues musician, sepia"}
},
"step_options": {"subtitles": {"animation": "karaoke"}}
}
Two doors, one account.
Plain HTTP/JSON with a key, or the same operations as MCP tools for an agent. Both spend the same balance and show up in the same run history.
Pay for what runs.
No subscription. Top up a balance by card, and each step draws on it at the model's price.
Free to begin
Every new account starts with a balance: enough for a first real project, not just a test.
$5.00 on sign-upCard, when you need it
Add funds by card through Stripe whenever you need to. Usage-billed models draw on the balance at the provider's rate.
$10 – $2,000Flat per generation
Image and video steps have a fixed price you see before you run.
Shown before every run