
A wall of text becomes parts
Paste a prompt and it splits into typed bloks you can name, move and test.
Web
Break any prompt into bloks, see what each one adds to the prompt you ship, and keep every version. When something breaks, find the blok that did it.


Paste a prompt and it splits into typed bloks you can name, move and test.

Each edit is saved. Compare any two versions, blok by blok.

Run on your own models and see which part changed the answer.
Something else? Write to support@soroucsh.com.
Yes. Every prompt and version is visible only to you. Nothing becomes public unless you create a share page, which needs Performance, and you can take that page down at any time.
They run your prompts on the provider you choose and nothing else. Keys are encrypted at rest, shown once when you save them, and never written to logs. The provider bills your own account. A model on your own computer or network runs straight from your browser, so its requests and its key never touch our servers.
Any model from 20+ hosted providers, such as OpenAI, Anthropic, Google, OpenRouter, Azure, Bedrock, Mistral and Groq; any OpenAI-compatible endpoint; or a model on your own machine through Ollama or LM Studio. Add as many as you like, each with its own label. Local models run from your browser: Chrome and Edge reach them anywhere on your network after asking once, Firefox reaches models on the same computer, and Safari needs the model served over https.
Everything you need to write, save, version, run once and copy a prompt, with no limits on prompts, bloks or versions. Testing, analysis, comparison and sharing are in Performance.
One typed part of a prompt: context, constraint, example or expects. Each blok owns an exact span of the compiled prompt. Expects bloks describe what a good answer does, and never appear in the prompt itself.
In your account, with every version. You can export them all from Settings at any time, as Markdown and JSON.