// Generated from the feature catalog
Right-Sized Intelligence
Picks the Right Brain for the Job · Shipped · workstation, home
Automatically routes simple queries to fast local models and reserves deep reasoning for complex system diffs.
Halbert does not route every turn to a single monolithic cloud model. It maintains dedicated tiers — Guide (fast conversational triage), Specialist (complex sysadmin synthesis and code diffing), and Vision (image and diagram analysis) — each backed by independently configurable endpoints.
Incoming messages are scored for intent and complexity by the intake pipeline. The TierRouter checks available hardware capabilities and configuration before selecting the optimal endpoint. Debug mode surfaces the routing decision for every turn so you can see exactly why a query went where it did.
Limits and invariants
Section titled “Limits and invariants”If a configured endpoint goes down, routing gracefully degrades to the next available tier rather than failing silently.
Where this lives
Section titled “Where this lives”halbert_core/halbert_core/model/tier_router.pyhalbert_core/halbert_core/model/llm_config.pyhalbert_core/halbert_core/intake/pipeline.py