Version control for your prompts

Skills are reusable prompt packages. Each skill has a SKILL.md file with frontmatter config (name, description, allowed tools, model settings). You can invoke skills via slash commands (/skill-name arguments), import/export them as ZIP files, and test them with the same test framework used for prompts.

Yes. The Agent Builder lets you create agents from 5 templates (Chat Assistant, Code Assistant, Support Agent, Research Agent, or blank). Configure 13 sections including model, tools, memory, guardrails, and sub-agents. Test with single-turn, multi-turn, and trajectory tests. Export to Python, LangGraph, Claude SDK, or OpenAI Assistants.

Three types: Single-turn (input to response), Multi-turn (conversation simulation), and Trajectory (agentic task with tool-call verification). You can import test challenges from JSON/CSV/Excel, run model comparisons across test suites, and track costs per test case.

Prompt Studio ships 9 built-in validators: Contains, NotContains, Equals, NotEquals, Regex, NotRegex, JSON Schema, LLM Judge (with model selector, custom criteria, and threshold), and Custom. Plus 64 evaluators from the marketplace. Evaluation modes: all must pass, any must pass, or weighted average.

Git is great for code, but it doesn't let you run and visualize prompt outputs. Prompt Studio combines the versioning of Git with a built-in testing lab, agent builder, and matrix model comparison.

Yes, but encrypted. API keys are stored with Fernet encryption on the backend. Keys never touch the browser and are decrypted only at runtime for LLM calls. Zero markup on token costs.

Always. One-click JSON or ZIP export for prompts, skills, and test cases. Export agents to Python, LangGraph, Claude SDK, or OpenAI Assistants. Import with conflict resolution (Skip/Overwrite/Merge). No vendor lock-in.

Yes! Early Access is completely free during Developer Preview. No credit card required. Create unlimited prompts, skills, and agents.

Yes. The web version includes full team support: multi-organization management with 4 role levels (Owner, Admin, Editor, Viewer), real-time document locking, and Slack notifications. The desktop app is single-user but can connect to the web version.

Any provider with an OpenAI-compatible API or LiteLLM support. Bring your own keys — zero markup on token costs. Run local models with Ollama.

Use [[folder/prompt-name]] to compose prompts from reusable blocks. You can even pass variable overrides: [[persona | tone=friendly]]. Circular dependencies are detected automatically, with a max depth of 5 levels.

Every save creates a version you can restore. On deploy, the entire prompt tree (including all injections) is snapshotted into a single immutable version with content hashing. Roll back to any deployment with one click—non-destructive rollback creates a new deployment.

Injections ([[...]]) pull in other prompts as reusable blocks—your DRY principle for prompts. Variables ({{...}}) are dynamic values filled at runtime with JS expressions, nested properties, and built-in helpers like $now and $uppercase().

Yes. Your prompts are your IP. API keys are Fernet-encrypted on the backend. Prompt Studio is GDPR compliant and hosted in Europe. Export all your data anytime.

The desktop app is local-first: everything runs on your machine with no account required. It has the latest features including the agent builder, skills, and the full testing lab. The web version includes all team features (RBAC, real-time collaboration, Slack notifications) but is currently behind on newer features like agents and skills. The desktop app can connect to the web version for syncing and team workflows.