AI Configuration Reference

AI transform definitions live in the transform YAML file, not in the top-level Renderflow config.

AI transform fields

Key	Required	Notes
name	Yes	human-readable transform name
ai	Yes	ollama or openai
model	Usually	model identifier passed to the provider
prompt	Usually	prompt template containing optional {input}
endpoint	No	overrides backend base URL
api_key	No	plaintext fallback, discouraged
api_key_env	No	environment variable name for key lookup
artifact_path	No	also save generated output to a file
cache_path	No	AI cache file path
prompt_version	No	extra cache-busting dimension
from / to	Yes	graph/source and target formats
cost / quality	Yes	planner metadata

Example: Ollama

transforms:
  - name: local-summary
    ai: ollama
    model: mistral
    prompt: "Summarize:

{input}"
    endpoint: http://localhost:11434
    from: markdown
    to: markdown
    cost: 1.5
    quality: 0.8

Example: OpenAI-compatible

transforms:
  - name: polish-copy
    ai: openai
    model: gpt-4o-mini
    prompt: "Rewrite for clarity:

{input}"
    api_key_env: OPENAI_API_KEY
    cache_path: .renderflow-ai-cache.json
    prompt_version: v2
    from: markdown
    to: markdown
    cost: 2.0
    quality: 0.9

Warning

Prefer api_key_env so credentials stay out of version-controlled YAML.