EvaluatePrompt templates

Prompt templates

⚠️
Preview — this feature is being built; the design below is the spec.

A prompt template is a named, versioned prompt with {{variables}}. Production code, evaluation runs and the Console all refer to the same version, so the prompt you tested is the prompt you ship.

Create and version

curl -sS https://app.syntic.ai/v1/prompts/templates \
  -H "Authorization: Bearer $SYNTIC_API_KEY" \
  -H "syntic-beta: prompt-tools-2026-12-01" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "support",
    "messages": [
      {"role": "system", "content": "You answer billing questions for Acme…"},
      {"role": "user", "content": "{{ticket}}"}
    ],
    "model": "Amara S1.2"
  }'

Each save creates a new immutable version (support-v13). Labels such as production and staging point at versions and can be moved: support@production.

Use a template in a request

{ "prompt": { "id": "support@production", "variables": { "ticket": "I was charged twice." } } }

prompt replaces messages (and model, unless you override it). The response includes prompt_version so logs show exactly which version answered.

Promote with evidence

  1. Save the candidate as a new version.
  2. Run it against support@production in an evaluation run.
  3. Move the production label only if it wins. Moving a label is instant and reversible.

Rules

  • Variables are required unless given a default in the template ({{tone|friendly}}).
  • Templates are workspace resources; keys from other workspaces cannot read them.
  • Deleting a version that a label points at is refused.
Was this page helpful?