Introduction
PromptVersions is a workspace for building, versioning, testing, and shipping LLM prompts — with Git-like control across 7 major LLM providers.
PromptVersions gives prompt engineering the same rigor software gets: every edit is an immutable version, every run is logged, and every prompt can be evaluated, deployed to environments, and called over an API — all from one beautiful workspace.
What you can do
- Version everything — each save is a new immutable version with a commit message; diff or restore any of them.
- Run across providers — stream responses token-by-token from OpenAI, Anthropic, Google, DeepSeek, xAI, Mistral, or OpenRouter.
- Evaluate — define metrics and test cases, then batch-eval a version or auto-eval on every save.
- Build chains — wire multiple prompts together, passing each step's output into the next.
- Ship to production — pin versions to
dev/staging/prod, run A/B tests, and call prompts over the inference API.
How it works in three steps
1
Write your prompt
Create a prompt with {{ variable }} placeholders inside a workspace folder.
2
Test & iterate
Run it against any provider in real time, compare models side-by-side, and refine.
3
Version & ship
Save a version with a commit message, deploy it to an environment, or call it via the API.
Before you start
PromptVersions is bring-your-own-key — you connect your own provider API keys, and your usage is billed by the provider directly. After signing up you'll:
1
Create an account
Sign up, then you'll land in your first workspace.
2
Add a provider key
Open the AI Providers panel and paste a key from any supported provider. Keys are encrypted at rest and never shown again.
3
Create your first prompt
Add a folder, create a prompt, and run it.
Your keys stay yours
Provider keys are encrypted with Fernet at rest and are write-only in the UI — once saved, they're never returned to the browser. See Provider Keys.