2/9/24 Updates
Recent Updates: New prompt editor, better graphs, Thin SDK & new CLI, and new OpenAI model versions
- New prompt editor lets you test multiple versions/models at once
- Better observability with faster, interactive graphs & personalized table views
- Thin SDK and new CLI = full developer control
- New OpenAI model versions –
0125&0125-previewSee all the details here
Experiment faster with new prompt editor
Previously Freeplay only let you edit one version of a prompt at a time. Now you can load multiple versions at once, swap models, test them side by side, and compare results right in the browser. And soon you’ll be able to save & load test cases for faster experimentation with your data.
Better, faster graphs on Sessions dashboard
LLM observability is a key part of using Freeplay, and the Sessions dashboard has some fresh updates to make it easier & faster to explore your data and find what you’re looking for. In the past 10 days we’ve made several changes:
- Interactive graphs filter results in the table - click any graph value to filter, then dig in
- Much faster load times – big performance update!
- Personalized table configuration – control column visibility & order to match your needs
Thin SDK & New CLI for Prompt Version Management
The vast majority of companies we work with aren’t using frameworks for their integrations, they’re writing their own code. They want as much control as possible for production software.
This is why we’ve introduced a significant update to our SDK that breaks each key Freeplay feature – prompt management, logging, CI-style test runs, etc. — down into separate methods. (We also do all this at the code level, without any proxy service, so there’s nothing to get in the way of your production traffic.) We offer the same level of support for Python, Node & Java/JVM languages.
Together with our new CLI for prompt bundling that lets you check out prompt versions from the Freeplay server into your source code, developers can have full control over things like custom routing logic to different models, handling complex fallback scenarios for error handling, calling any model, etc.
Plus: New OpenAI models!
New OpenAI model versions are live -- you can now try 0125/0125-preview in the Freeplay app.