Welcome to Prompt Assay

Prompt Assay is a version-controlled workbench for building, refining, testing, and shipping LLM prompts. It runs entirely on your provider API keys. We don't resell tokens, we don't mediate your LLM traffic, and we don't see your spend. You bring Anthropic, OpenAI, and/or Google keys, and Prompt Assay gives you the editor, version history, AI-assisted critique, evaluation suites, and collaboration tools around them.

Who it's for

• Prompt engineers shipping production prompts and needing version control, diffing, and rollback.
• Teams sharing a library of prompts across a workspace with role-based access.
• Researchers running evaluation suites and judge-scored comparisons across models.
• Developers who want a REST API + TypeScript SDK to pull prompts at runtime.

What you get

• Full-featured editor with syntax highlighting, real-time token budget, cost estimation, and a 10+2 lint-rule set (ten always-on + two model-aware) for prompt quality issues.
• Version control with full history, diff viewer, restore, and branching from any prior version.
• Six AI pair panels: Brainstorm, Critique, Improve, Rewrite, Compare, Convert · all powered by your BYOK keys. The eval-suite Judge and regression sweep call the same workbench model. Convert turns a working prompt into a complete Agent Skill bundle in one shot (BYOK-only, preview-before-apply).
• Evaluation suites with test cases, rubrics, and automated judge-based scoring across six dimensions. Plus regression testing on version change, comparing version N vs N+1 side-by-side.
• Playground for live runs, variable substitution, and a Compare mode that runs two versions of the same prompt against the same model.
• Reusable fragments with variable slots for composing prompts.
• Public REST API + TypeScript SDK for fetching prompts at runtime.
• Team workspaces with role-based access, invitations, and per-workspace BYOK keys.

Five concepts in two minutes

Prompt

The atomic unit of work. Has a title, content, prompt type (system / user / multi-turn / template), optional target model, and intent metadata.

Version

An immutable snapshot of a prompt's content at a point in time. Every edit creates a new version with a change source tag (manual / ai-improve / restore / branch / etc.).

Workspace

A container for prompts, folders, tags, fragments, BYOK keys, and billing. Every user has a personal workspace automatically; team workspaces are manually created and support collaboration.

BYOK

Bring Your Own Key. Per-workspace API keys for Anthropic, OpenAI, and Google. Required for the playground, AI pair, evaluation judges, and any LLM-backed feature.

Tier

Your workspace's billing plan: free, solo, team, or enterprise. Tier controls member caps, monthly LLM call quota, and public API access. Each workspace has its own tier.

Next steps

1. Read Core concepts for the one-page mental model.
2. Follow Create your first prompt to get something into the editor.
3. Configure your first BYOK key to unlock playground + AI features.
4. Skim Keyboard shortcuts so you move fast from day one.