Routing agent
The cheapest model
that actually works.
route reads each prompt and dispatches it to the weakest model that clears your quality bar, escalating only when the margin demands it. Same answers, a fraction of the invoice.
route · quality-floor dispatch · per-request ledger
Why teams install
What route earns back
Cuts LLM spend 40–70%
Measured on a replay of your own traffic, not a vendor benchmark.
‹c9d3a84›Quality floor is a contract
Drop below your rubric and route escalates mid-stream, automatically.
‹e5b1f90›Failover that fails over
Provider outages reroute in one RTT. PagerDuty stays quiet.
‹1fa7c62›How it works
Three steps. No faith required.
01
Score
Every prompt gets a complexity read and a risk margin in microseconds.
02
Dispatch
The weakest sufficient model answers; hard prompts climb the ladder.
03
Audit
Per-request ledger of model, cost, and margin — reconciled to your invoice.
Proof band
One packet. Every claim accounted for.
packet rig/route · request 91f3c02 prompt "summarize this 402-line diff" scored complexity 0.31 · margin +0.12 served local-8b · priced after a short conversation vs priced after a short conversation frontier floor PASS — rubric ≥ 0.85 held themis SEALED — ledger reconciled to invoice
Every route run emits a packet like this — hash-chained, verifier-signed, and queryable from your own dashboard. How ProofPackets work →
Buyers
Who runs route
Platform Lead
AI SaaS, 40M req/day
Head of Eng
usage-based startup
SRE
high-traffic consumer app
Before you install
The obvious objections
Q Why did the metered rate change?
Repriced against real fleet economics — the old rate assumed frontier-heavy routing that most traffic never hits. a rate set after a short conversation reflects what teams actually pay once the router is doing its job. Pro priced after a short conversation adds private endpoints and committed-volume pinning on top.
Q How does route decide “weakest sufficient model” without silently degrading quality?
Every prompt gets scored against your quality rubric before dispatch — drop below the floor and route escalates mid-stream, automatically. The per-request ledger shows the model, the cost, and the margin, so a quality regression shows up in the ledger before it shows up in a support ticket.
Q What happens during a provider outage?
Failover reroutes in one RTT to the next model on the ladder that still clears the rubric floor — no manual intervention, no PagerDuty page. The ledger records the failover event so you can reconcile it against the invoice after the fact.
Pricing
Start free. Pay when it pays.
Generalist ships today. Domain Expert adds vertical memory + priors. Custom is built to your stack.
Free
Priced after a short conversation
100K tokens a month routed. Full ledger access.
Usage
Priced after a short conversation/1K tok
Pay per token routed. All providers, all ladders.
Pro
Priced after a short conversation
Private endpoints, fleet pinning, committed volume.
Runs well with
