Skip to content

AI Advisor

The AI Advisor is an optional feature that reviews your diagrams and suggests improvements. It runs in the background and surfaces hints as small badges on nodes.

The advisor evaluates your model against C4 best practices:

  • Missing edges — nodes that should be connected but aren’t
  • Incorrect shapes — a database container with the default rectangle instead of a cylinder
  • External system flags — third-party services that should be marked external
  • Description quality — vague or missing descriptions
  • Flow coverage — steps without linked processes, or processes not covered by any flow
  • Structural issues — frontend-to-database shortcuts, duplicate nodes, unclear boundaries

When the advisor finds something, a colored dot appears on the affected node:

  • 🟠 Orange dot — a warning that likely needs attention
  • 🟢 Teal dot — informational suggestion

Select the node to see the hint message in the Properties panel. Some hints have actions — like “Set shape to cylinder” or “Mark as external” — that you can apply with one click.

If a hint isn’t relevant, check the dismiss box in the Properties panel. Dismissed hints won’t reappear for that node unless the model changes significantly.

Open Settings (gear icon in the top bar) and go to the AI Review tab.

Choose a provider and enter your API key:

ProviderModels
OpenAIGPT-4o, GPT-4, etc.
AnthropicClaude Sonnet, Claude Opus, etc.
GoogleGemini Pro, etc.
GroqMixtral, Llama, etc.
MistralMistral Large, etc.
DeepSeekDeepSeek Chat, etc.
OllamaAny locally running model

Click Fetch models to see available models for your provider. Select one and save.

The advisor runs automatically after a 10-second pause following model changes. It doesn’t run on every keystroke — only when the model’s structure changes (adding/removing nodes, changing edges, etc.).

The advisor is purely optional. Scryer works fine without it — it’s just an extra layer of review.

When using cloud providers, your model’s structure (node names, descriptions, edges) is sent to the API. No source code is sent. If you prefer to keep everything local, use Ollama with a locally running model.