AI Tools Strumento aggiornato 3 giorni fa

AI Workflow Builder

Design a multi-step AI workflow visually. Each step has a name, a prompt template with {{variable}} placeholders, a target model, and a temperature. Branch steps into parallel arms, merge them back, reorder them, and export the workflow as versioned JSON or readable Markdown. This is a planner/designer - it does not execute the workflow.

Sempre gratuito Facile da usare Risultati immediati Riservato e sicuro

Utilizzo AI Workflow Builder

AI Workflow Builder

Design multi-step AI workflows visually. Add steps, pick models from the live catalog, branch and merge, then export as JSON or Markdown. Designer only — it never executes the workflow.

Flow diagram

Visual representation of the workflow. Click a node to jump to its editor.
Add a step to start building the diagram.

Workflow steps

Reorder with the arrows, duplicate, or remove. Branch with the next-step chips.
No steps yet. Click Add step or pick a template.

Variables across workflow

Every {{variable}} placeholder detected. Provide values for the preview run.
No variables detected. Use {{name}} in any prompt template.

Preview run

Resolves every prompt with your values. Designer only — no model is called.
Resolved prompts appear here.

Export JSON

Link agli strumenti

Copiato negli appunti.

Aiutaci a migliorare questo strumento

Leggi il contenuto

Scopri come funziona lo strumento e quando utilizzarlo.

How to Use This Tool

Build a workflow from step cards, watch the flow diagram redraw as you edit, and export the finished design. Everything runs in your browser; nothing is executed and nothing is sent to a server.

1. Add steps

Click Add step to create a new step. Give it a unique name, write the prompt template, and use {{variable}} placeholders for runtime inputs. Reorder steps with the arrows or drag them, duplicate a step, or remove it with the trash icon. Changes autosave to your browser storage.

2. Pick a model

Open the model picker to search the live models.dev catalog, grouped by provider. Featured chips give one-click access to popular models. Models that do not support temperature disable the slider automatically. If the catalog is unreachable, a small offline list is used instead.

3. Branch and merge

Use the Next steps picker on a step to route it to one or more following steps. Multiple successors create branches; steps with multiple incoming edges are merge points. The flow diagram draws arrows, branch arms, and merges automatically, and flags cycles.

4. Review with a preview run

Fill the Variables across workflow values once, then run the preview. Every step resolves its prompt with your values (including {{stepName.output}} references to earlier steps), and the linter reports unknown variables, unclosed braces, duplicate names, and cycles.

5. Export

Export the workflow as versioned JSON ({ version, exportedAt, title, steps }) and import it back later, or copy it as Markdown for pasting into a team doc. JSON import accepts either a bare array of steps or the wrapped object shape.

Features

  • Branching DAG with merge points, not just a linear chain
  • Live SVG flow diagram that redraws as you edit; click a node to jump to its editor
  • Live model catalog from models.dev with search, provider grouping, and featured chips
  • Per-step temperature slider (0-2) that disables for models with fixed sampling
  • Variables detected per step with {{name}} regex; step-output references via {{stepName.output}}
  • Preview run that resolves every prompt with your variable values and lints the workflow
  • Per-step and total token estimates
  • Versioned JSON export plus readable Markdown export
  • Startup templates (linear chain, classifier branch, summarizer chain)
  • Validation with inline errors and undo/redo
  • Autosave to localStorage under awb-workflow

Domande frequenti

Risposte concise alle domande che gli utenti si pongono prima di fidarsi del risultato.

No. It is a planner and designer only. It never calls an LLM API or runs the steps. Export the JSON and execute it in your own pipeline.

No. Everything runs in your browser. The only network request is the models.dev catalog, fetched to build the model picker; your workflow is never uploaded.

From the public models.dev dataset (https://models.dev/models.json), cached in your browser for 24 hours. If it cannot be reached, a bundled offline list is used instead.

Any {{name}} in a prompt template is an input variable. You provide its value in the Variables panel for the preview. A {{stepName.output}} reference pulls the output of an earlier step, resolved by step name.

Import accepts either a bare array of step objects or the wrapped { version, exportedAt, title, steps } shape, as long as each step has at least a name and prompt.

In your browser localStorage under the key awb-workflow, saved automatically as you type. Clear your browser data or hit Clear to reset.

Commenti (0)

Per discussioni tra utenti, casi limite e consigli aggiuntivi.
Lascia un commento
Il tuo commento sarà visibile a tutti una volta inviato.
Non ci sono ancora commenti. Lascia il primo commento!
Facci sapere!

Segnala un problema con questo strumento

Descrivi il problema che hai riscontrato, in modo che possiamo analizzarlo e migliorare lo strumento.

Ciò che aiuta di più
Includi i dati inseriti, il risultato atteso, il risultato effettivo, il browser/dispositivo utilizzato e, se utile, uno screenshot.
Il pulsante "Screenshot" acquisisce la pagina nel browser e allega l'immagine generata a questo modulo.
Screenshot a tutta pagina
Realizzato nel tuo browser con html2canvas.
Non è stato ancora acquisito alcun screenshot.
Strumento Relazione