mtlm-router / pricing
The runtime, trunk, and tools are open source (mtlm-router, Apache-2.0). What is licensed are lane packs — calibrated decision heads plus their eval evidence and provenance. You are not buying a weight file; you are buying the certification around it: taxonomy design, leak-free eval, operating point, and recalibration support.
Every pack in the catalog — certified and demo — is Apache-2.0. They exist as proof, not product: the format, the verification tooling (lane_verify), and what a certified eval looks like, checkable by anyone on public data. Includes lanes that fail certification shipped openly — that is the point of the catalog.
Send your route table (names + a few phrasings each). You get a ~7KB .head trained and calibrated on your corpus — eval report, abstention threshold, one free re-train after a week of shadow traffic. 3 business days.
Open the intake form or emailA certified pack under a commercial-use license for a named domain, plus a 12-month recalibration window: you monitor drift with tools/monitor.py, and when the operating point needs refitting we recalibrate against a fresh labeled split — the guarantee tracks your traffic, not the training set. Production use on your own infrastructure, no egress, no per-call metering.
The same /v1/route on api.mtlm-router.intrane.fr, metered by peage (EUR, Stripe top-ups, signed receipt per charge). New wallets start with free credit. Shared evaluation environment — latency and capacity are not representative of a dedicated deployment.
A single-tenant instance on bare metal — your own endpoint, every catalog lane installed, no per-call metering. Production posture; the metered API above is for trying it first. Get an instance
source of truth: mtlm-heads/PRICING.md