SAGE

Reflective agent demo for schema-driven Howgrams JSON editing.

Edit Diagram

Schema validation enabled
Bookmark: /sage/94c4c184-523f-43a8-a4b4-96cbaa1c78aa/
Ask for changes; the agent drafts JSON, validates it, and retries if invalid.

Live Preview

Real-time rendering of your diagram JSON

Preview will appear here as you edit

SAGE, up close

Structured artifacts a generic LLM can’t reliably produce out of the box.

SAGE is our agentic building block for creating and editing custom, structured artifacts — the kind of proprietary IP that demands correctness. Users describe intent in plain language; the agent returns the full artifact, not a vague suggestion.

i.

Draft

The agentic loop turns a natural language request into a first candidate artifact.

ii.

Validate

Strict typing, invariants, and registries define what “valid” means for your domain.

iii.

Reflect

Validation feedback converts fuzzy language into reliable, testable structure.

iv.

Revise

The loop repeats until every constraint is satisfied — repeatable, deterministic results.

Example artifact: Howgrams — interactive explainer diagrams for docs, onboarding, training, and system walkthroughs.

How it works

Agentic APIs for schema-driven editing.

The pattern

  1. Draft: the model proposes a full artifact (e.g., a Howgrams diagram JSON document).
  2. Validate: validate the output against your schema and invariants.
  3. Reflect: if validation fails, feed the error messages back to the model.
  4. Revise: the model outputs a corrected full artifact.

Why this works