Intelligence that ships,
not intelligence that presents.

We embed with your team, build the model your product actually needs, and stay until it runs without us.

Inference cost−41%
$4,180
per million calls, last 30 days
Weekly runs1,204
Mon → Sun
Latency p95
184 ms
across 6 regions
This week
38experiments
12shipped
4rolled back
0incidents
Accepted
92.4%
Routing
Ranking
Fallback
Review queue
+7
3 checks waiting on a human, none older than a day
Traffic split
64 / 22 / 14
A · B · held out
Coverage
99.98%
uptime, rolling quarter
Drift watchclear
21 days, one flag raised
Inference cost−41%
$4,180
per million calls, last 30 days
Weekly runs1,204
Mon → Sun
Latency p95
184 ms
across 6 regions
This week
38experiments
12shipped
4rolled back
0incidents
Accepted
92.4%
Routing
Ranking
Fallback
Review queue
+7
3 checks waiting on a human, none older than a day
Traffic split
64 / 22 / 14
A · B · held out
Coverage
99.98%
uptime, rolling quarter
Drift watchclear
21 days, one flag raised
Inference cost−41%
$4,180
per million calls, last 30 days
Weekly runs1,204
Mon → Sun
Latency p95
184 ms
across 6 regions
This week
38experiments
12shipped
4rolled back
0incidents
Accepted
92.4%
Routing
Ranking
Fallback
Review queue
+7
3 checks waiting on a human, none older than a day
Traffic split
64 / 22 / 14
A · B · held out
Coverage
99.98%
uptime, rolling quarter
Drift watchclear
21 days, one flag raised
Inference cost−41%
$4,180
per million calls, last 30 days
Weekly runs1,204
Mon → Sun
Latency p95
184 ms
across 6 regions
This week
38experiments
12shipped
4rolled back
0incidents
Accepted
92.4%
Routing
Ranking
Fallback
Review queue
+7
3 checks waiting on a human, none older than a day
Traffic split
64 / 22 / 14
A · B · held out
Coverage
99.98%
uptime, rolling quarter
Drift watchclear
21 days, one flag raised
Inference cost−41%
$4,180
per million calls, last 30 days
Weekly runs1,204
Mon → Sun
Latency p95
184 ms
across 6 regions
This week
38experiments
12shipped
4rolled back
0incidents
Accepted
92.4%
Routing
Ranking
Fallback
Review queue
+7
3 checks waiting on a human, none older than a day
Traffic split
64 / 22 / 14
A · B · held out

4.9 average — 38 teams shipped in the last twelve months

The teams already running on it
NorvidCirquaHaldenPeleusOrbikVarnelLumekSevrinKaltroMerundOkvistTrellinByvorAnsorQuorin
NorvidCirquaHaldenPeleusOrbikVarnelLumekSevrinKaltroMerundOkvistTrellinByvorAnsorQuorin
Pricing

You pay for decisions served.
Not for seats watching them.

Live · a Scale account, this billing period
963,200 / 1,500,000 decisions
184 msp95, 6 regions
99.98%uptime, rolling quarter
0incidents this period
Starter

For the first model you put in front of users.

$360/month
billed monthly $4,320 a year
  • 200,000 decisions a month
  • 3 models · 1 region
  • p95 under 250 ms
  • Nightly drift watch
  • $1.40 per extra 1,000 decisions

Month to month · cancel any time

Start on Starter
ScaleMost teams

For a team shipping to production every week.

$1,200/month
billed monthly $14,400 a year
  • 1,500,000 decisions a month
  • 25 models · 6 regions
  • p95 under 200 ms
  • Drift watch and a human review queue
  • Two engineers embedded, 4 days a month
  • $0.90 per extra 1,000 decisions

Moves up without a migration

Start on Scale
Estate

For every model you run, watched every night.

$3,600/month
billed monthly $43,200 a year
  • 6,000,000 decisions a month
  • Unlimited models · every region
  • Latency target written into the contract
  • The runbook handed back on the last day
  • $0.60 per extra 1,000 decisions

Signed on your paper

Talk to an engineer
About us

We put two engineers in your team,
find the one number a model can move,
and ship it before the quarter ends.

In production
120+
models in production,
watched every night
100%
of engagements handed back with the runbook

“They left us something we could operate on a Tuesday without calling anyone.”

+9 Data platform teamSeries B logistics
Every day
520k+
decisions served every day across our live systems
The bench
20+
engineers who have run this in production before