Prompt management, reimagined

Git-like control
for your AI prompts.

PromptVersions brings Git-like version control and prompt versioning to your AI prompts — version, test, evaluate, and deploy across every major LLM provider from one beautiful workspace.

Start for freeSee how it works
Bring your own API keysCancel anytime
promptversions — code-review.prompt
PROMPT
System: You are an expert code reviewer.
User: Review the following for issues:
{{ code_snippet }}
Run
gpt-4o · OpenAI
OUTPUT
## Code Review Summary
✦ Bug: line 23 — null ref before `.length`
Add null check: `if (!arr) return [];`
✦ Tip: use `const` for immutable bindings
✦ Perf: memoize the sort comparator
Overall quality ████████░░ Good
Features

Everything you need for
serious prompt engineering

A complete toolkit for building, versioning, and running prompts across every major LLM provider.

Version History
Every prompt edit is a new immutable version with commit messages, just like Git. Diff any two versions side-by-side.
v3 → v4
- Summarize this article in 3 bullet points.
+ Summarize this article in 3 concise bullet points.
+ Keep each under 15 words. Preserve technical terms.
Multi-step Chains
Chain prompts into ordered steps that reference each other's output, versioned and run end-to-end with live SSE streaming.
ExtractSummarizeTranslate
Multi-provider
Switch between every major LLM provider without changing your workflow. One interface for all.
openaianthropicgoogle+4 more
Evals & Test Cases
Score outputs with seven metric types — keyword match, JSON validity, regex, and LLM-as-judge — across saved test cases.
✓ keyword_match
✓ json_valid
✗ regex_match
2 / 3 passed
AI Prompt Optimizer
Analyze a prompt across seven criteria and get an AI-rewritten version that preserves your template variables.
Clarity 6/109/10
Environments
Deploy pinned versions to dev, staging, and prod slots — ship prompt changes without touching application code.
devstagingprod · v12
A/B Traffic Testing
Split production traffic between two versions by percentage and compare results straight from real run logs.
Variant A 70%Variant B 30%
Team Collaboration
Invite teammates with owner, editor, or viewer roles — invitations require acceptance, access is enforced on every route.
Avaowner
Sameditor
Priyaviewer · pending
API & Webhooks
Call any deployed prompt or chain with a scoped API key, and get signed webhooks on deploys, A/B stops, and eval runs.
POST /inference
version.deployed → 200
Skill Registry
Version multi-file agent skills (SKILL.md plus reference docs), promoted through draft/stable/live and fetched by your agent code via API.
GET /inference/skills/:id?version=live
{ "files": [{"path": "SKILL.md"}] } → 200
How it works

From idea to version in three steps

1
01
Write your prompt
Create prompts with template variables like `{{ variable }}`. Organize them in workspace folders.
2
02
Test & iterate
Run against any provider in real-time. Watch the response stream in, tweak, and rerun instantly.
3
03
Version & share
Save a new version with a commit message. Diff any two versions. Restore any past state.

Supported providers

One workspace.
Every major LLM.

OpenAI
GPT-4.1, GPT-4o, o1, o3-mini
Anthropic
Claude Opus 4.7, Sonnet 4.6, Haiku 4.5
Google
Gemini 2.5 Pro, Flash
DeepSeek
DeepSeek Chat, DeepSeek Reasoner
xAI
Grok models
Mistral
Mistral Large, Mistral Small
OpenRouter
100+ models, one key
Pricing

Simple, usage-based pricing

Start free. Upgrade when you need more runs, workspaces, and concurrency.

View pricing & plans
Starts free · upgrade anytime

What is prompt versioning?

Prompt versioning is version control for AI prompts. Every time you change a prompt, PromptVersions saves an immutable, diffable version — the same way Git records a commit — capturing the exact content along with the model, temperature, and settings it ran with. That history lets you compare any two versions, reproduce past behavior, and roll back the moment a change regresses in production.

Prompts are code that ships to users, but they rarely get the same discipline as code. Without prompt version control, a one-line edit can silently change outputs for everyone, with no diff, no attribution, and no way back. PromptVersions gives prompts a proper workflow — versions and diffs, side-by-side provider comparison, evaluation metrics, A/B tests, and deployments to dev, staging, and prod — across OpenAI, Anthropic, Google, DeepSeek, xAI, Mistral, and OpenRouter, all from one workspace.

New to the topic? Start with our complete guide to prompt versioning, or browse all prompt engineering guides.

Frequently asked questions

What is prompt versioning?

Prompt versioning is version control for AI prompts: every time you edit a prompt you save an immutable, diffable version — like a Git commit — so you can compare changes, reproduce past behavior, and roll back safely. PromptVersions applies this Git-like workflow to prompts across every major LLM provider.

How is PromptVersions different from storing prompts in Git?

Git tracks files; PromptVersions tracks prompts as first-class objects with the model, temperature, and settings each version ran with. You get diffing, one-click restore, side-by-side provider comparison, evals, A/B tests, deployments to dev/staging/prod, and an inference API — none of which a raw Git repo gives you.

Which LLM providers does PromptVersions support?

OpenAI, Anthropic, Google, DeepSeek, xAI, Mistral, and OpenRouter — from one workspace. You bring your own API keys, so you only ever pay your providers directly.

Can teams collaborate on prompts?

Yes. Workspaces support owner/editor/viewer roles, invitations, shared version history, in-app notifications, and per-member usage analytics, so a whole team can version and ship prompts together.

Is PromptVersions free to start?

Yes — there's a free plan, and because you bring your own provider keys there are no per-run platform fees. Paid plans add the optimizer, evals, team seats, usage logs, and API access.

Ready to ship better
prompts, faster?

Join developers who use PromptVersions to version, test, and perfect their AI prompts.

Get started for free
No credit card required