Platform · FinOps for AI

Every token, allocated.

AI is the fastest-growing, least-governed line on your bill — and provider dashboards stop at the API key. Optmin connects every input and output token to the model that served it, the team that called it and the product that paid for it, in the same FOCUS fabric as your cloud cost.

Every provider Token-level allocation Unit economics
AI token spendall providersToken spend by provider · input + outputOpenAIAnthropicBedrockVertexinputoutput
100%
AI spend allocated
Per-token
attribution
Real-time
anomaly guardrails

Token-level visibility, business-level allocation

Input, output and cached tokens broken out per request — then aggregated by model, key, user, team or any custom dimension. The number lands with the people who can act on it.

  • Token-in / token-out per request
  • Allocate every API key to a team or product
  • Map shared keys with rules you control
Token allocation100% mappedALLOCATED BY TEAM · 184 keysΔweb-eng38%▲ 5%platform24%▼ 3%ml-infra20%▲ 1%support18%▼ 2%100% of AI spend allocated

One schema across every provider

OpenAI, Anthropic, Google Gemini, AWS Bedrock, Azure OpenAI, Mistral and Cohere unified into a single model — folded into the same cost fabric as AWS, Azure and GCP. Read-only ingest, live in minutes.

  • Multi-provider, one normalized schema
  • Same fabric as your cloud bill
  • No agents, no pipelines, no code changes
Tokens by providermeteredGroup by: Provider ▾Last 30 days ▾AI PROVIDERtokensOpenAI42M▲ 8%Anthropic31M▼ 3%Bedrock24M▲ 2%tokens mapped to team · last 30 days97M tok

Guardrails that act before the invoice

ML-driven anomaly detection catches token spikes and prompt-length blowouts in near real time — and CloudFlow can pause a key or swap to a cheaper model automatically, inside your policy.

  • Real-time spike & prompt-blowout detection
  • Budgets & alerts per team, product or model
  • Signals → actions: pause key, swap model
Guardrailsautodoc-summarize · prompt v6 · +41%⏸ Pause key↻ Swap model✓ Saved $1.8k

Unit economics for AI

Tie tokens to value: cost per inference, per user, per feature — the margin numbers product and finance actually want, in the same showback and chargeback engine as your cloud cost.

  • Cost per inference, per user, per feature
  • Margin visibility on AI-powered products
  • Showback & chargeback through DataHub
Unit economicsvalueCost / customer$0.42Cost / feature$1.9kGross margin73%Margin on AI-powered featuresEvery token tied to a customer & outcome
Challenges we solve

AI bills don't tell you who used what.

Provider dashboards stop at the API key, shared costs land in one lump, and finance can't tie any of it to the value it creates.

The challenge

Provider dashboards stop at the API key

You can see total spend, but not which team, product or customer drove it.

How OPTMIN solves it

We attribute every token to a team, product and customer — the same way you allocate cloud cost.

The challenge

Shared AI costs can't be split

Eval pipelines, embeddings and internal copilots get billed in one indivisible lump.

How OPTMIN solves it

Split platform-wide AI cost across teams with fixed, proportional or custom rules you define.

The challenge

A runaway prompt 10×s spend overnight

A bad prompt or a loop can burn budget for days before it ever hits the invoice.

How OPTMIN solves it

ML-driven anomaly detection flags token spikes in near real time and can pause a key or swap a model automatically.

The challenge

Finance can't price the product

Without unit economics, margins on AI features are a guess.

How OPTMIN solves it

Cost per inference, per user and per feature — live, in the same engine as your cloud chargeback.

Everything to run AI responsibly

From the first API key to enterprise scale.

The pieces FinOps teams actually need when AI spend starts to matter.

Provider connections
Read-only key or billing-file ingest. Usage flows in within minutes.
Token accounting
Input, output and cached tokens reported per request, any dimension.
Multi-provider schema
One model across OpenAI, Anthropic, Gemini, Bedrock, Azure OpenAI and more.
Model-level analysis
Cost per model, per workload, per request — find rightsizing the way you do for compute.
Allocations
Build allocations from key, model, prompt metadata or custom tags.
Shared cost split
Distribute eval pipelines and internal copilots with logic you control.
Anomaly detection
ML-driven detection of token spikes and prompt-length blowouts.
Automated guardrails
Turn signals into actions — pause a key or swap a model, no human in the loop.
Unit economics & chargeback
Cost per inference and per feature, tied to revenue through DataHub.

Stop guessing what AI is costing you

See your token spend mapped to teams, products and outcomes. Book a walkthrough.