Model Routing Principles
How Canoma decides response behavior in Auto and pinned modes, without exposing internal implementation details.
This page explains user-visible routing behavior in Canoma. It focuses on what you can expect, not on internal execution details.
What Influences Routing
Canoma adapts response behavior using:
- Query intent (triage, synthesis, remediation, or lookup)
- Requested output format (brief, checklist, timeline, or plan)
- Scope constraints (assets, timeframe, severity, and confidence thresholds)
- Active workspace policies and safety controls
Auto Mode Behavior
In Auto mode, Canoma continuously balances:
- Answer quality for the task
- Response speed for the workflow stage
- Stability under changing load or external dependency issues
You may notice small style differences between runs when Auto mode optimizes for these factors.
Pinned Mode Behavior
When you pin a model mode in Settings, Canoma prioritizes that choice while still applying standard safety and policy controls.
Use pinned mode when you need predictable tone and output structure across repeated workflows.
Resilience Expectations
If a preferred option is temporarily unavailable, Canoma attempts a compatible option so workflows can continue.
What this means for users:
- Responses may vary slightly in tone or depth.
- Structured output requests are still honored when possible.
- Re-running the same query after transient issues typically returns to normal behavior.
Skills and Routing
Skills run only when the active model path supports skill execution for the task.
Practical guidance:
- Keep a fallback query that does not depend on custom skills.
- Validate critical skill workflows after major workspace configuration changes.
- For high-stakes outputs, require explicit structure and verification fields in the prompt.
When To Use This Page
Use this page to:
- Explain why response style can change between Auto and pinned modes.
- Set team expectations about consistency vs adaptability.
- Debug workflow-level behavior without relying on internal architecture details.