Skip to content

Platform · Routing

Model Routing & Spend

How PURE routes work across model tiers, meters AI usage per principal, and routes inbound leads to nurture plans. Live config from pure_routing_rules + pure_rate_limits. Snapshot: 2026-06-26.

Model tiers

TierUsed forModel class
LightDrafting, intake refine, classification, in-page assistsHaiku class
StandardSummaries, routing decisions, everyday agent workSonnet class
HeavyDeep build, review, and reasoning-critical lanesOpus class

Tiers are chosen per task by pure_routing_rules (model routing) so light work never pays for a heavy model. Every call is metered in audit_log.

AI rate & spend caps (per principal)

Calls-per-minute rate & a daily spend ceiling

“Spent today” is the running tally against each principal’s ceiling.

PrincipalCalls / minDaily $ capSpent today
mike (superadmin)600$500$0.10
lane-a (Design)200$50$0.00
lane-b (Platform)200$50$0.00
verifier120$10$0.00
client60$5$0.00

Every MCP call is metered in audit_log; the health-scan cron escalates anomalies. Enabling caps = enforce pure_rate_limits in the MCP guard.

Lead routing plans

First matching rule wins

Inbound leads match the first rule (by intent · source · type) and enter that nurture plan. Higher intent routes faster.

#PlanIntentMatchAction
1#21 High Intent SellerhighsellerDrive to a listing appointment
2#29 High Intent BuyerhighbuyerFast cadence + auto saved-search
3#28 Specific Home InterestanyOpen House · buyerProperty-match follow-up
4#30 Re-Engaged BuyeranySaved Search Alert · past-clientWarm-back sequence
5#26 New Lead AuditmidanyVerify + route before nurture
6#NUR Long-Term NurturelowanySlow drip until intent rises

Rules live in pure_routing_rules (ordered); two auto-create a saved search on match. Edit order/intent there to re-tune routing.

Built by Code from live config — this page reflects the tables as configured; no figures invented.