AI cost management for OpenAI, Anthropic, and cloud-billed models, on one ledger with your cloud

What it covers

Five sources, two ways in. The direct APIs connect with an admin key that CloudQuell uses only to read usage and cost reports; cloud-billed AI arrives with the cloud integration you already have.

OpenAI API

Daily cost per model from the Costs API, joined to token counts from the Usage API and attributed per project. Up to 12 months backfilled on connect.

Anthropic API

Daily Claude cost by model, workspace, and token type — uncached input, cache writes, cache reads, and output — with a cache-read share you can narrow to one workspace. Up to 12 months backfilled on connect.

Amazon Bedrock

Read from your AWS Cost and Usage Report: by model, token type, routing, linked account, region, and tag.

Azure AI Foundry

Read from your Azure bill: Azure OpenAI, the other models Microsoft sells through Foundry, and Claude bought through Azure Marketplace — by model, model developer, token type, deployment type, subscription, and region. Provisioned throughput (PTU) and Marketplace Claude show as their own buckets.

Gemini / Vertex AI

Read from your Google Cloud billing export: Gemini, Claude and other Model Garden models, Imagen, Veo, embeddings, and grounding, by project. Gemini 1.x / PaLM-era usage is billed by characters, so it shows dollars without token counts.

Counted once

Cloud-billed AI stays inside cloud spend. AI Spend is a read-only lens over those rows, and its All AI total is never added to cloud totals — so the Overview, budgets, and allocation still reconcile to your invoices.

Where each AI dollar comes from

Every source lands in the same view with a team boundary of its own. Whether it already sits inside a cloud total decides how it is counted.

How it connects

Read-only reporting calls on every source: two admin keys for the direct APIs, nothing new for the clouds.

Direct APIs

An OpenAI Admin API key with Read Only permissions (sk-admin-…) and an Anthropic Console admin key (sk-ant-admin01-…). Each is validated live on connect and stored in AWS Secrets Manager, never in the database.

Cloud-billed AI

Nothing to connect. The Bedrock, Azure AI Foundry, and Gemini / Vertex buttons appear in AI Spend on their own once your AWS, Azure, or Google Cloud bill carries that spend.

No code changes

CloudQuell reads the providers' billing and usage reports, so nothing sits in your request path. Team attribution comes from the OpenAI projects and Anthropic workspaces you set up in each provider console.

Freshness

Daily, not real-time. OpenAI and Anthropic sync daily and re-pull a trailing three-day window; cloud billing data settles a day or two behind them, so an All AI total is dated by its oldest source.

What's out of scope

Request tracing: cost per prompt, per feature, or per end customer is finer than either billing API reaches. Per-key dollars: API keys show token and request counts only. Seat subscriptions such as ChatGPT Team or Claude Max are not billed per API call.

Plans

AI Spend, cloud-billed sources included, is on every plan — Free covers up to $10K of monthly tracked spend and three connected integrations. Cost centers, allocation rules, budgets, alert rules, and MCP access are on paid plans: Starter $99/month up to $50K, Growth $199/month up to $200K.

What it unlocks

With every AI dollar in one place, it can be owned, budgeted, and watched like the rest of the bill.

• Team attribution

  Route each OpenAI project and Anthropic workspace to a cost center with a project_id or workspace_id tag rule — the same cost centers your cloud spend uses, so a team's AI line sits next to its AWS line. Cloud-billed AI allocates with that cloud's tag rules.

  AI cost allocation by team →

• Budgets and alerts per team

  Scope a budget to a cost center, to specific OpenAI projects or Anthropic workspaces, or to a tag, with thresholds at 80% and 100% by default. Spend Threshold, Trend Spike, and Anomaly Scan rules post to Slack, Microsoft Teams, or email.

  AI spend budget alerts →

• Anomalies per model

  On OpenAI and Anthropic, Anomaly Scan compares each model with its own baseline, so a runaway loop on one model is flagged even when the team total looks ordinary. Bedrock is watched as a service on your AWS bill.

  Cost anomaly detection →

• Effective $ per 1M tokens

  A blended rate that divides only the spend that bought tokens. Provisioned capacity, guardrails, and server tools stay in the total but out of the rate, with a note on how much of the spend the rate covers.

  Compare model prices →

• Agent access over MCP

  Ask Claude, Cursor, or any MCP client for OpenAI spend by project or Claude spend per workspace. get_ai_spend leaves cloud-billed sources out unless asked, and reports direct and cloud-billed totals separately so an agent can tell which part is already in cloud spend.

  MCP server (docs) →

Common questions

What does an AI cost management platform do?

It collects what you pay for AI — API invoices and cloud-billed model usage — attributes it to the teams that spent it, and puts budgets and alerts on it: FinOps for AI spend. CloudQuell does this from billing data: OpenAI through a read-only admin key and Anthropic through a Console admin key used only for its usage and cost reports, and Bedrock, Azure AI Foundry, and Gemini / Vertex AI from the cloud bills you already connect, in one AI Spend view beside your cloud and Snowflake spend.

Can I track OpenAI, Claude, and Bedrock costs in one place?

Yes. With more than one AI source, AI Spend opens on All AI with a per-source split, and Claude on Bedrock and Claude direct share one model line. Bedrock, Azure AI Foundry, and Vertex AI stay counted inside cloud spend; the All AI figure is analytical and is never added to cloud totals, budgets, or allocation.

Does CloudQuell show LLM cost per prompt or per request?

No. It reads the providers' billing and usage reports, which stop at model, project or workspace, token type, and day. Cost per prompt, per feature, or per end customer needs an LLM observability tool that instruments the call site; plenty of teams run both. API keys show token and request counts only — neither provider reports dollars per key.

How do I allocate AI spend to teams?

Give each team its own OpenAI project or Anthropic workspace and issue its keys there. In CloudQuell, add a tag rule per team — project_id = proj_… or workspace_id = wrkspc_… — on that team's cost center. Projects or workspaces with no rule, and everything in the Default project or Default workspace, show as Unallocated. Cost centers and allocation rules are on paid plans.

Is AI cost management the same as AI spend management for SaaS seats?

No. AI spend management in the procurement sense tracks seat licenses, SaaS subscriptions, and purchase approvals. CloudQuell tracks the metered bill — API tokens and cloud-billed model usage. Seat subscriptions such as ChatGPT Plus, Team, and Enterprise or Claude Pro, Max, Team, and Enterprise aren't billed per API call and don't appear in either provider's cost report, so they are out of scope.

How much does AI cost tracking cost?

AI Spend is on the Free plan for up to $10K of monthly tracked spend and three connected integrations. Budgets, alerts, allocation, and MCP access are on paid plans — Starter $99/month up to $50K of monthly spend, Growth $199/month up to $200K — flat, never a percentage of your bill, with a 14-day trial and no credit card.