Model Transparency Panel
The Model Transparency Panel gives you full visibility into which AI models are handling each stage of your task — and lets you override them when you want more control.
Opening the panel
Section titled “Opening the panel”Access the Model Transparency Panel via the persistent icon in the workspace header, or use the keyboard shortcut ⌘ + Shift + M.
What the panel shows
Section titled “What the panel shows”The panel displays a live view of the current pipeline state:
- Active stage — which pipeline stage is currently running (highlighted with an amber indicator)
- Model per stage — the AI model assigned to each of the six stages
- Assignment type — whether each model was automatically routed or manually overridden
- Token count — tokens consumed per stage for the current session
- Estimated cost — cost per stage based on token usage
Stage layout
Section titled “Stage layout”Each of the six stages is displayed as a row:
| Stage | Assigned Model | Type | Tokens | Cost |
|---|---|---|---|---|
| Orchestration | Claude Opus 4 | Auto | 2,340 | $0.07 |
| Planning | Claude Sonnet 4 | Auto | 1,890 | $0.02 |
| Building | GPT-4o | Manual | 8,420 | $0.05 |
| UI Generation | Gemini 2.5 Pro | Auto | 4,210 | $0.03 |
| Debugging | Claude Sonnet 4 | Auto | — | — |
| Review | Claude Opus 4 | Auto | — | — |
Stages that haven’t run yet in the current task show dashes for token and cost columns.
Manual model override
Section titled “Manual model override”You can override the automatically assigned model for any stage:
- Click the dropdown arrow next to any stage’s model name
- Select an alternative model from the list of available models
- The stage label updates to show “Manual” instead of “Auto”
Visual indicators
Section titled “Visual indicators”- Auto-assigned stages — model name in standard text, “Auto” badge
- Manually overridden stages — model name with an amber underline, “Manual” badge in amber
Resetting overrides
Section titled “Resetting overrides”Click the Reset all button at the bottom of the panel to return every stage to automatic model routing. You can also reset individual stages by opening the dropdown and selecting “Auto (recommended).”
Model display names
Section titled “Model display names”Codewick shows friendly model names rather than technical API identifiers:
| What you see | What it means |
|---|---|
| Claude Opus 4 | Anthropic’s most capable reasoning model |
| Claude Sonnet 4 | Fast, high-quality coding model |
| GPT-4o | OpenAI’s flagship multimodal model |
| Gemini 2.5 Pro | Google’s advanced reasoning model |
Hover over any model name to see a one-sentence tooltip describing what that model excels at.
How routing decisions are made
Section titled “How routing decisions are made”When set to automatic, Codewick’s routing algorithm considers:
- Task type — what kind of work this stage needs to do
- Priority mode — your current setting (Cost, Quality, Speed, or Balanced)
- Token cost — how expensive each model is per token
- Model benchmarks — how each model performs on relevant tasks
- Availability — whether the model is currently accessible
The algorithm updates automatically as new models become available — you don’t need to do anything.
When to use manual overrides
Section titled “When to use manual overrides”Manual overrides are useful when:
- You know a specific model handles your programming language particularly well
- You want to try a more powerful model for a complex task
- You’re debugging a tricky issue and want the strongest reasoning model available
- You want to reduce costs on a simple task by selecting a lighter model