What is Prompt Hub?
Prompt Hub is a centralized repository for teams to manage, iterate, and deploy prompt templates. It’s a collaborative workspace to version control templates and use them across playground, tasks, APIs, and experiments.Access our Prompt Hub API reference here

Why Use the Arize Prompt Hub?
The Arize Prompt Hub transforms prompt management from scattered code snippets into a rigorous engineering discipline. It serves as a centralized, search-friendly repository for your entire ecosystem of prompts, allowing you to track their evolution through immutable version control.- Unified Prompt Ecosystem
Stop hunting through slack messages or codebase comments. Store every “Prompt Object”, including template variables, model parameters, and tool definitions, in a single, searchable repository that tracks past, present, and future versions of your assets. - Immutable Version Control
Treat prompts with the same rigor as code. Every refinement creates an immutable, hashed version, allowing you to safely rollback changes, reproduce past results, and tag specific versions (e.g., prod-v1.2) for deployment without ambiguity. - Rapid Experimentation & Comparison
Close the iteration loop by loading any prompt version directly into the Playground. Test changes against real-world spans or golden datasets, and use side-by-side comparison to visually diff outputs and verify improvements before merging. - Compatible, Easy-to-Use SDK
Seamlessly push and load prompts in code, allowing easy use within your applications. - Automated Prompt Optimization
Move beyond manual tweaking. Optimize underperforming prompts into to let the system leverage feedback loops and meta-prompting, automatically generating and ranking better variations for you.
What is a Prompt Object?
At Arize, we expand the definition of a prompt to a Prompt Object, which consists of everything a complete prompt entails.- Prompt Template - system, user, and assistant messages that make up the LLM’s context
- Invocation Parameters - temperature, frequency penalty, top p/k, etc.
- Tools/Functions - specific APIs, programmatic functions, or outside functionality
- Response Format (aka output schema) -> configure specific JSON output
