AI Model Router
The AI Model Router is Learn Domains' abstraction that selects, invokes, and logs language models based on task type, quality needs, and cost constraints.
Also known as: model-router · llm-router
Why it matters
Hard-coding one model everywhere destroys margin or quality. Routing lets you classify with small models, generate long-form with premium models, and enforce consistent token and cost accounting for 70–80% gross margin targets.
How it works
Services declare task intent, classification, summary, draft, strategic analysis. The router maps intent to the right model, records usage against your credits, and respects your balance before running a job.
Common mistakes
- Bypassing the router with direct API calls in feature code.
- Using premium models for trivial yes/no classification.
- Skipping cost logs, making margin dashboards blind.
- Changing models in env vars without updating registry documentation.
Best practices
- Route every AI call through the router, no exceptions.
- Default to cheaper models unless QA requires premium output.
- Alert when task cost drifts above registry estimates.
- Review usage breakdown on the Credits page monthly.
Learn Domains perspective
Not every task needs the same depth, classifying a question is lighter work than writing a 2,000-word draft. Learn Domains meters AI through Mission Fuel so heavy generation costs more than a quick analyst question, you see usage on the Credits page without picking models yourself.
FAQ
- Can I choose the model per request?
- Learn Domains routes by task type for consistent quality and predictable credit use.
- Are model costs visible to users?
- Credits abstract the cost, see breakdowns and forecasts on your Account → Credits page.
- What happens if a model provider fails?
- Router fallbacks apply per configuration; failed jobs release credit reservations.
Next steps
- 1Review Credits usage breakdown by category.
- 2Identify tasks burning premium credits unnecessarily.
- 3Align team workflow to start with structured Content Operations drafts before optional AI enhancement.
Knowledge graph
Parent terms
Related concepts
rag · digital-asset-intelligence · content-operations