PromptVault
Ship prompt changes without touching your codebase
Start new thread
Yevhenii Ozden

PromptVault - Ship prompt changes without touching your codebase

by•
If you're building an LLM app, your prompts are constantly changing - and every change shouldn't require a PR, CI run, and a full redeploy. PromptVault is a prompt management tool for AI builders. Edit prompts live in production, version every change, and roll back in one click. Connect Once → Draft → Live - without touching your codebase. Built for indie developers, AI founders, and product teams shipping LLM features.

Add a comment

Replies

Best
Yevhenii Ozden
Maker
📌
Hey PH 👋 Built this after watching our team open PRs just to fix a comma in a system prompt. Every time we wanted a tone tweak, our engineer had to stop what they were doing. It added up fast. PromptVault started as an internal tool — then we realized every team building with LLMs had the same problem. Would love to hear how you're currently managing prompts in production. Git? Hardcoded strings? A Google Doc somewhere? 👀 Happy to answer any questions about the product, the stack, or where we're taking it next. Check our web and try PromptVault with a free plan!
Karim Ben

How do you handle secrets and variables inside prompts, like API keys or user specific fields, without exposing them in the UI?

Artem Litvinenko
Maker

@karimbenkeroum Hi Karim, thank you for the question!
Technical Co-Founder here :)

Variables in prompts are lexical placeholders, resolved at runtime, not stored.

The consequence for your question: a user-specific field or API key is never stored in the prompt.

The app stores 'Hello {{api_key}}', and whoever calls the SDK injects the real value at request time. The secret lives in the consumer's environment, not in PromptVault.

To sum it up, regarding "how do you avoid exposing it in the UI" - the value never enters the system. The editor does show the token literally ({{api_key}}), but that's just the placeholder name, not a value.

Ethan Brooks

This hits a real pain — I've definitely opened a PR just to tweak the wording of a system prompt 😅

The flip side I keep thinking about: the whole pitch is "edit live without a redeploy," but that also means a bad prompt edit can hit production instantly with no CI in between. How do you guard against that? Is there any staging/preview step, or a way to run a prompt change against a set of test cases before it goes Live?

Rollback in one click helps, but I'd want to catch the regression before users do. Congrats on the launch 🚀

Artem Litvinenko
Maker

@ethanbrooks_capdrop Hi Ethan, thank you for the question and congratulations!
Technical Co-Founder here :)

Yes, for teams that need it - you can set specific version you need to "staging" and update it to "live" when you are ready!

Regarding test cases - prompt regression testing is on our roadmap