Latitude is an open-source platform for AI prompt engineering, deployment, and evaluation. It helps teams ship reliable AI features by closing the loop between prompt development, production traces, and continuous improvement.
With Latitude, you can:
- Design and version prompts collaboratively
- Test iterations in an interactive playground
- Evaluate and improve prompts systematically
- Deploy prompts as API endpoints
- Monitor performance with automatic tracing tied to real usage
Key Features at a Glance
- Prompt Manager: Create, version, and collaborate on prompts with a powerful editor supporting advanced features like variables, conditionals, and loops through PromptL
- Playground: Test prompts interactively with different inputs, parameters, and tool configurations
- AI Gateway: Deploy prompts as API endpoints that stay up-to-date with published changes
- Datasets: Manage test data for batch evaluations and regression testing
- Evaluations: Assess prompt performance via LLM-as-judge, programmatic rules, or human review
- Run experiments: Use datasets and associtated evaluations to run batch experiments over 2 or more prompts variations.
- Traces & Observability: Automatically capture all interactions with prompts and models
- Integrations: Seamlessly integrate with your existing stack via SDKs and APIs
The Reliability Loop
Latitude is built around a reliability loop that turns real usage into improvements:- Design: Create and iterate on prompts using the Prompt Manager
- Test: Validate behavior in the Playground with test inputs
- Deploy: Publish prompts as endpoints through the AI Gateway or render them locally with the SDKs
- Trace: Automatically capture production interactions with telemetry
- Evaluate: Assess performance using various evaluation methods
- Improve: Refine prompts and repeat the loop as you learn from traces