Versioned Prompt Library
Every prompt change creates a new version. Roll back, compare, and promote versions — never lose a prompt that was working.
Floating Labels
Assign labels like production, staging, or experiment to any prompt version. Your app calls the label, not a specific version — so you deploy by moving the label, not by changing code.
Variable Templating
Use {{variable}} syntax in prompts. At runtime, pass values and the template renders correctly — reuse the same prompt structure across many contexts.
GitHub Bidirectional Sync
Prompts sync to a GitHub repo automatically. Edit in the UI → commit appears in Git. Merge a PR → prompt pulls into TotalCtrl. Treat prompts like code.
Evaluation Datasets
Define input/output test cases for any prompt. Run evals to check that a new version still produces the right outputs before promoting to production.
LLM-as-Judge Scoring
Each eval run is scored 0–100 by an LLM judge using your defined criteria. See score trends across versions and catch regressions before they reach production.
Runtime API
Fetch any prompt at runtime via GET /api/v1/prompts/{lib}/{slug}?label=production. Responses are cached for 60 seconds — low latency, always current.
Multi-model Metadata
Store preferred model, temperature, and other parameters per prompt version. Switch models without hunting through code.