Prompt Lab Guide: Version, A/B Test, Track, and Perfect Your AI Prompts

# Prompt Lab Guide: Version, A/B Test, Track, and Perfect Your AI Prompts

Guesswork is the enemy of good AI output. You tweak a system prompt, get a better response, and immediately wonder: was it the wording, the structure, or just a lucky draw? Without a way to measure the difference, you are stuck iterating by feel.

Prompt Lab is Vincony's dedicated prompt-engineering workspace. It combines four tightly integrated panels — Versions, A/B Test, Experiments, and Library — into a single hub for everyone from casual users on the Free plan to power users on Pro, Power, or Business. This guide walks through every panel and shows you exactly how to move from a raw prompt idea to a tested, documented, reusable asset.

---

The Four Panels at a Glance

Navigate to Prompt Lab and you will see four tabs across the top of the interface. Each one covers a different stage of the prompt lifecycle.

---

Panel 1 — Versions: Git for Your Prompts

The Versions tab is the heart of Prompt Lab. Every change you save becomes a numbered version with a full diff from the previous one, so you never lose a working iteration.

What you can do here

• Create a new version — go to the New Version sub-tab, paste your prompt text, add an optional label (e.g. "added tone instructions"), comma-separated Tags, and a Notes field explaining your reasoning. Hit Save Version or press Ctrl+Enter.
• Live diff — while typing in the editor the interface shows a live inline diff against the previous version so you can see exactly what changed before committing.
• AI optimization — click Optimize with AI to have the platform rewrite your current prompt for effectiveness. Review the suggestion, then save it as a new version.
• Compare two versions — select any two versions using their checkboxes, then click Compare Selected. A resizable split pane appears with both prompts side by side, plus an expandable Inline diff section below.
• Fork — create an independent branch from any version. Useful when you want to explore a structural change without disrupting your main lineage.
• Restore — load any version back into the editor with one click. The content populates the New Version tab so you can adjust and save a new iteration.
• Star — mark your best versions so they appear in the Starred sub-tab for quick access.
• Pin — pins float to the top of your version list so reference versions are always visible.
• Test — open any version in a Test Prompt dialog to run it against a live model and optionally save the output as a new version.

View modes

Switch between three view layouts using the icons in the top-right of the versions list:

• List view — the default card layout with full action menus.
• Timeline view — a visual branch diagram showing how versions relate to each other chronologically.
• Changelog view — a structured log format you can copy and paste into documentation or share with a team.

Tags and search

Use the tag filter chips to narrow down versions by label (e.g. "tone", "system-prompt", "v2"). The search bar supports full-text search across prompt content and version labels. Press / to jump straight to the search field.

Export and share

The Export button downloads all visible versions as JSON. The Export and Share popover lets you push them directly to Google Drive, Google Docs, Notion, Dropbox, or OneDrive, or send them via email — useful for teams sharing prompt libraries across tools.

---

Panel 2 — A/B Test: Fast Variant Generation

The A/B Test panel takes a different approach. Instead of manually writing each variant, you describe a prompt and the AI generates multiple rewritten versions, scores them, and declares a winner.

This panel is available on the Pro plan and above (Free and Starter accounts see a plan upgrade prompt).

Step-by-step: running your first A/B test

Step 1 — Open the A/B Test tab. Go to Prompt Lab and click the A/B test tab.

Step 2 — Set your optimization goal. Choose one of three goals using the goal selector: - Quality — maximizes output accuracy and depth. - Speed — optimizes for brevity and faster responses. - Cost — finds the most efficient phrasing to reduce token usage.

Step 3 — Choose how many variations to generate. Select 3, 5, or 7 variations. Three is a good starting point. Five or seven give you more surface area but cost more credits (the tool costs 3 credits per run).

Step 4 — Enter your prompt. Paste the prompt you want to optimize. You can also pick a quick-start preset — Content prompt, Chatbot prompt, or Analysis prompt — to pre-fill a starter phrase.

Step 5 — Run the test. Click the main submit button. The AI generates each variation, produces a sample response for every variant, and scores each one across four dimensions: Quality, Relevance, Creativity, and Conciseness (all on a 1–10 scale).

Step 6 — Review results and the winner. Below the output you will see a highlighted Winner card showing which variation performed best for your chosen goal and a brief explanation of why.

Step 7 — Follow up. Use the follow-up action buttons to go deeper without retyping: - Test with different goal — re-run the winner optimized for a different dimension. - Expand winning variation — add more context and detail. - Merge best parts — combine strengths from multiple variants into one. - Rewrite for different audience — get a technical version and a plain-language version simultaneously.

Step 8 — Save or compare. Copy the winning variation, save it to favorites, or take it back to the Versions tab to start a new version lineage from the winner.

---

Panel 3 — Experiments: Multi-Model Live Testing

The Experiments panel (the Experiments tab, previously the standalone "Prompt Tracker" tool) is for users who want to run a real prompt against multiple AI models simultaneously and compare the results with hard data.

How experiments work

Create a named experiment using the + button in the sidebar. Each experiment is a persistent container for all the test runs you run under that name.

Select which models to include. The available options include Gemini 3 Flash, Gemini 3 Pro, GPT-5 Mini, and GPT-5 Nano. Toggle as many as you want — each selected model runs your prompt in parallel.

Enter your prompt in the composer and click Run Test. The platform fires requests to every selected model simultaneously, records the response, and measures latency in milliseconds.

After the run completes, each result appears in a results table with:

• Model — which model produced the response.
• Latency — how long it took in milliseconds.
• Cost — credit cost for the run.
• Quality score — automatically scored out of 10 by the platform's scoring engine.
• Relevance score — also auto-scored out of 10.
• Manual rating — click the star row to add your own 1–5 star rating on top of the auto-scores.

Expand any row to read the full response text and the AI's reasoning behind the quality score.

The Leaderboard

The right-hand sidebar aggregates all runs for the selected experiment into a Leaderboard. Cards rank each model by average quality, average relevance, average latency, total credit cost, and an efficiency metric (quality per credit). The top-ranked model gets a trophy icon.

This makes it easy to answer practical questions: is the faster, cheaper Gemini 3 Flash close enough in quality to justify using it over Pro? The leaderboard answers that with data, not intuition.

Exporting results

Click CSV to download all experiment runs as a spreadsheet. The export includes every column — model, prompt text, label, latency, cost, scores, manual rating, and timestamp — ready for analysis in Excel or similar tools.

Use /models to explore the full model catalog on Vincony before deciding which ones to include in your experiments.

---

Panel 4 — Library: Your Prompt Collection

The Library tab (previously the standalone "Prompts" tool) is a personal bookmark manager for prompts you want to keep and reuse.

Prompts land here whenever you save one from the chat interface or any tool. Each saved prompt shows its title, a preview of the content, tags, and the date it was saved.

• Search — full-text search across title, content, and tags. Press / to jump to the search bar instantly.
• Tag filtering — click any tag chip to filter to prompts with that label. Click again to clear.
• Use in Chat — click the chat icon on any prompt to open it directly in Vincony Chat as the starting message.
• Copy — copy the prompt to clipboard with one click.
• Export — download your entire library (or your current filtered view) as a Markdown file.

Think of the Library as the destination for prompts that have graduated from experimentation. Once you have run a prompt through Versions and Experiments and you know it works, save it to the Library so it is always one click away.

---

Tips for Getting the Most from Prompt Lab

Tag everything consistently. Tags are the connective tissue across the Versions and Library panels. A tagging convention like "system-prompt", "user-facing", "draft", and "production" makes filtering fast as your collection grows.

Use labels to document your reasoning. The version label field ("added tone instructions", "removed hedge language") is a commit message for your prompt. Your future self will thank you.

Let the Experiments panel inform model choice, not habit. It is easy to default to the same model out of familiarity. Running even three or four experiment runs on a real task gives you actual latency and quality data to make an informed decision. Pair this with the model comparison tool for a broader view.

Combine A/B Test and Versions. A common workflow: run an A/B test to quickly surface strong structural variants, then take the winner into the Versions tab and iterate on the wording incrementally. This gives you the speed of AI-generated variation with the precision of fine-grained version control.

Keyboard shortcuts save time. Press ? anywhere in the Versions panel to see all available shortcuts. The most useful: n to jump to New Version, / to focus the search bar, and Ctrl+Enter to save.

---

Plan Availability

The Versions, Experiments, and Library panels are available to all signed-in users. The A/B Test panel requires the Pro plan or higher due to the higher credit usage per run.

If you are on the Free or Starter plan and want to test variations today, use the Versions panel to write and save two or three variants manually, then use Compare Selected to see them side by side. It is a more hands-on approach that works well for smaller iteration cycles. Upgrade to Pro or above when you are ready for automated scoring and multi-model experiments.

---

Start Measuring Your Prompts

The difference between a prompt that kind of works and one that reliably delivers is usually a few carefully tracked iterations. Prompt Lab gives you the tooling to find that difference — version history that never loses your work, A/B testing that surfaces winners in seconds, experiment data that tells you which model to trust, and a library that keeps your best prompts at hand.

Open Prompt Lab and save your first version. Everything else follows from there.