▶modelweave.io/console/landing⌘K

Replay 30 days of LLM traffic against any rule. Ship the rule that wins the audit.

ModelWeave is a routing console for backend SREs running multi-LLM stacks. Toggle a rule and the scatter rebinds in place; the right rail tells you the dollar delta with a Bayesian Thompson 95% credible interval — cited to arXiv:2402.02563, not a vendor blog.

▶ open /replay⇆ /router$ pricing
K1.7

Counterfactual Replay Engine

Toggle a rule. The 30-day scatter rebinds via View Transitions. The pill prints the savings before you write the PR.

K1.5

Bayesian Thompson Router

Beta-Bernoulli posterior per provider arm. Every counterfactual ships with a 95% credible interval, sourced to arXiv:2402.02563.

K3.2

Edge-KV Stream

The right rail reads decisions from Vercel Edge KV at 5ms p99. The live feel is not theatre — it is sub-frame latency.

persona · backend sre · 10K-1M req/day

You operate in tmux, Datadog, Grafana, Honeycomb, PagerDuty. You audit every dependency you ship. You do not trust black-box “AI recommended” routers. This is the one that hands you the math.