Measure every AI dollar. Then cut the waste.

trAIce meters every LLM call and every AI tool your team uses, attributes each dollar to the feature, customer, that customer’s users, team, and agent behind it, then hands you the specific change that lowers the bill: cost-validated model swaps, cache savings, and waste alerts you can act on.

Start freeRead docs

No credit card · open-source SDK · 5-minute install

AI spend measured
$3,182
Waste identified
$612

Live demo data · last 30 days

Helpdesk AI
Dashboard
Last 30 days
Last 30 days · Team plan
Daily spendUpdated now
30 days agoToday
Total spend
$3,182
30-day cost
Avg cost / call
$0.042
75,761 events
Tokens
7.8M
prompt + output
Top feature
chat-resolver
by spend
Top customer
Initech
by spend
The problem

AI spend is growing faster than the systems used to manage it.

Engineering sees model spend. Finance sees invoices. IT sees SaaS commitments. CTOs and VP R&D leaders still cannot point to the customer, feature, team, or employee that changed the budget.

Which customers are profitable on AI?

Your dashboard says “OpenAI: $4,012/mo.” That’s not an answer. You need cost per customer, joined to the revenue they pay.

Your bill jumps 40% and no one can explain why

A new feature? A chatty customer? A retry loop? A model you forgot to downgrade? Aggregate spend hides the leak.

Which team or employee pushed AI spend over budget?

Seat commitments, metered tools, and raw API usage land in different systems. Leaders need one team-level view and alerts before a burst becomes the monthly baseline.

The build side

What AI costs to make.

Start from your application’s context: feature, customer, and run. Do not start from a billing export. That’s the difference between “how much did the API cost” and “which feature is unprofitable.”

Per-customer margin

Cost vs revenue, per account.

Every call tagged to a customer and its users, joined to the revenue they pay.

Revenue$99
AI cost$410
Net margin −$311 · Initech
Cost-validated swaps
gpt-4ogpt-4o-mini
−68% cost96% equivalence30 samples
Prompt-cache savings
41%
hit rate · $2.7k saved
Agent economics

Cost per run, step, and retry.

run$0.21tool$0.09retry×3$0.27
Waste detection
  • Retry cascade $184
  • Runaway agent loop $132
  • Duplicate spend $96
New
The buy side

What your team spends using AI.

Beyond what AI costs to build, see employee usage from Codex, Claude Code, and internal sources you send through the API. Compare teams day by day, combine metered usage with seat commitments, and alert when an employee crosses budget.

Internal Spend
Employee AI Spend
Track employee tool usage separately from product AI spend.
Last 35 days
Daily employee AI spendUpdated now
30 days agoToday
Usage cost
$8.7k
Token estimate
$9.1k
Monthly commitment
$4.2k
Employees
47
Billable tokens
126M
EmployeeBillable est.Token est.
Liam O'Connor
Engineering · Codex
18.4M
$684
Aisha Rahman
Data & AI · Claude Code
9.7M
$318
Maya Chen
Engineering · Codex
7.9M
$276
Interactive demo

Three operating views in the real platform UI.

Each workspace opens with fresh fictional data, current alerts, and read-only controls.

$50K
Initech AI margin

Customer AI-margin leak

Northstar Helpdesk AI finds Initech driving more resolver cost than its customer revenue safely supports.

131
employees in one workspace

Internal AI spend abuse

The same Northstar workspace shows 131 people across 12 startup teams, with department, role, source, budget, and anomaly evidence.

1 workspace
one operating workspace

Product + employee AI spend

Northstar gives leaders one Team workspace for customer-facing AI economics, verified savings, and internal employee usage.

Control & roadmap

From measurement to the actions that cut spend.

We surface each saving with evidence and you apply it. Some levers you can turn on to act automatically. Nothing in your stack changes without your say, and we stay out of your request path.

Cache insights and opt-in exact cacheLive
Cost-validated model swapsLive
Waste detection: retries, loops, duplicate spendLive
Budget alerts at 80% and 100%Live
Block or downgrade a call over budgetComing soon
Routing and fallback enforcementComing soon
Idle-seat waste detectionComing soon
How we fit

One platform for both sides of company AI spend.

Gateways route. Evals score. Observability traces. Spend tools count seats. None of them connect what you build to what you buy.

What you wanttrAIceAI gatewayPortkey, LiteLLMEvalsBraintrust, LangSmithObservabilityDatadog, Helicone
AI cost per customer / tenantNoNoPartial
AI gross margin (cost vs revenue)NoNoNo
Cost per agent run / tool stepPartialNoPartial
Cost-validated model swapsNoPartialNo
Prompt-cache hit rate & savingsPartialNoPartial
Employee / team AI tool spendNoNoNo
Both build-side and buy-side AI costNoNoNo

Why not my AI gateway?

Portkey, LiteLLM, OpenRouter

A gateway routes traffic and can total your spend, then stops. We sit beside it as the meter and the optimizer: attribute every dollar to the feature, customer, that customer’s users, and agent, then turn it into savings. Routing is a different job, and these pair well with us. We are deliberately not in your request path.

Why not LLM observability?

Datadog, Helicone

Observability shows you the number and the trace, then leaves the “now what” to you. We pair the measurement with the levers that bring cost down: cache insights, cost-validated model swaps, and waste caught in loops and retries.

Why not an evals tool?

LangSmith, Langfuse

Eval tools tell you whether a change is good, not what it costs. When we surface a cheaper model you see the quality evidence and the cost delta side by side, so safe and cheaper are one decision.

Why not my provider’s dashboard?

OpenAI, Anthropic

One provider, one page, after the bill lands. We unify every provider, attribute spend to the feature, customer, user, and agent that drove it, and alert in real time as a cost starts climbing.

Why not tag spend in my cloud cost tool?

AWS, GCP tags

Cloud cost tools bucket by service or account. “OpenAI: $4,012/mo” is not an answer. We start from your app’s context (feature, customer, run) so you see which feature is unprofitable, not just how much the API cost.

Buy-side spend tools (Torii, Zylo) and AI-FinOps (Amnic, Finout) cover half of this. None tie it back to the product you ship, or turn it into a specific change that lowers the bill.

Your data

Built to measure your spend, not hold your prompts.

You get the full product on metadata alone. Prompt content is opt-in, and trAIce never sits in your request path.

Metadata by default

We run on tokens, cost, latency, and the tags you attach (feature, customer, user, run). Raw prompts and outputs are never ingested unless you explicitly turn capture on.

Prompts stay opt-in

Prompt capture is off by default. When you enable it, samples carry a 14-day TTL and are deleted immediately if you opt out.

Never in your request path

Events are sent after your LLM call, so trAIce cannot slow it down or take it offline. Under 1 ms local SDK overhead per tracked call at p99.

See the benchmark

Keys and certification

trAIce API keys are stored as SHA-256 hashes; provider keys you connect for replay are encrypted with AES-256-GCM. We are not yet SOC 2 certified: Type I is on our near-term roadmap, and we will walk your security team through our controls today.

Read the full detail on our security and data handling page.

Built for

Three people, three jobs, one platform.

CTO / VP R&D

Own the full AI operating budget.

See product economics and employee usage together, with accountable teams, active budget alerts, and a clear path from anomaly to action.

Engineering manager

Know which team, workflow, or person changed spend.

Daily team breakdowns, employee budgets, and deep-linked logs make cost reviews concrete without turning managers into billing analysts.

AI platform lead

Optimize with evidence, not blanket limits.

Trace cost to customer, feature, agent run, employee, and team. Validate model changes and apply targeted budgets where they matter.

“Provider invoices tell you how much. We built trAIce to tell you which customer, feature, and employee, so you can act before the bill arrives.”

The trAIce team

Frequently asked

Good questions, straight answers.

See where your AI budget really goes.

Open the demo with no signup. See a customer margin leak, an employee budget spike, and a unified product + employee workspace.

Start freeRead docs

50,000 events/month free, no credit card.