Revenue lies.Margin doesn't

Name: Nozle
Brand: Nozle

Every customer costs you something different. Every model reprices you on Tuesdays. See margin per customer the day it moves — not the quarter it ends you.

Book a teardown

Start free

Four things every AI company needs to ship.

Margin attribution, model-swap-safe pricing, hot-path usage caps, and a pricing engine live in 60 minutes — on one fabric, wired in once.

Margin, by name. Live.

OpenAI and Anthropic calls auto-captured; Gemini, Replicate or any other provider costed from your own rates. Each one attributed to the customer who triggered it. Underwater accounts surface in minutes — not at month-end.

Provider COGS joined to customer_id. Not a CSV. Not a quarterly reconcile.
Alerts the hour a customer flips negative. By plan, by cohort, by feature.
Re-rate the past against any future plan. "What if last month was on next month's pricing?"

margin ledger · livetoday · 14:32

customermrr7d trendgross margin

Northwind AI
Scale
$2,400
+471%
Helio Labs
Pro
$890
+158%
Quill & Co.
Pro
$890
−644%
Drift Studio
Starter
$190
−31-18%
Forge Robotics
Scale
$3,100
+266%
Pico Health
Pro
$890
−58-42%

2 underwater · 1 watchingcogs joined

model swap · plan v172 swaps · re-costed

planroute · was → nowimpact

Starter
tier · 01
gpt-4o-mini→gemini-1.5-flash
+12%
margin
Pro
tier · 02
gpt-4o→claude-sonnet-4
+8%
margin
Scale
tier · 03
claude-opus-4→claude-opus-4
—
no change
Enterprise
tier · 04
gpt-4o→gpt-4o
🔒
contract

versioned · reversiblere-costed from your cost sheet

Switch models rewriting your pricing.without

OpenAI cuts 50%. Anthropic ships smarter at 2× the tokens. Gemini lands. Your price lives on the plan; your cost lives in the Cost Sheet. Swap models and every new call is re-costed, per customer and per model.

Margin by model, per customer. Different customers, different models, same ledger.
Plans are versioned and reversible. Roll back v17 without a migration.
Replay any month against any plan. Forecast the swap before you ship it.

Cap a customer mid-run. No PRs.

An agent runs for hours. A power user 100×s the median. A free trial loops. Cap the spend at the gate — per-customer, per-plan, per-action — without an engineer hardcoding a rate-limit at 2am.

Enforcement on the hot path. Atomic check-and-deduct. No double-spend.
Cap mid-run, bill the outcome. Long agents stop being a margin event.
Per-customer override on the fly. Lift a cap for an enterprise. No deploy.

sdk.can() · enforcementhot path · /can

requestcostdecision

✓cus_8a2· gpt-4o · 12,400 tok
$0.249
allowed
✓cus_3kf· embed-3l · 84,000 tok
$0.017
allowed
✗cus_91q· agent.run · step 42
$5.04
capped
└─ plan: pro · max est. cost $5.00 · denied
✓cus_v0t· vector_read · 1,204
$0.024
allowed
✓cus_lt7· gpt-4o · 240,000 tok
$4.80
override
└─ ent_acme +$50 · 2pm · logged & escalated

3 allowed · 1 capped · 1 override0 PRs · 0 deploys

plan.compose · v18 draftno deploy

prompt · plain english

3 lines

>Charge $50/mo plus $0.002 per GPT-4o input token

>and $0.006 per output token. First 10k free.

>Annual customers get 20% off.

plan · compiled

json · valid

base$50.00/ month
meter[in]$0.002/ 1k tok · free_tier: 10,000
meter[out]$0.006/ 1k tok
discount20%annual_term

v18 · liveversioned · reversible

Ship a new pricing model in 60 minutes.

Per-token, per-action, hybrid, credits, outcome-based. Describe the plan in English; ship it versioned, without a deploy. The pricing page is no longer the slowest part of your release.

Compose any pricing shape. Token, call, second, action, outcome — combine per plan.
Versioned at the customer level. Grandfathering is a flag, not a migration.
A/B prices, then keep what wins. No re-platform. Just promote the version.

Not an AI startup?