For the head of IT who owns the AI bill

See your AI spend
with its evidence.

Cloud AI spend and agent governance for IT teams. Review supported provider costs, record subscriptions and set budgets for requests routed through Token Observe.

Start free 30-day trialExplore a sample workspace
Cloud pilot30 days free from workspace creation. No card or automatic paid conversion.

Synthetic sample · no account or keys

Who owns this charge?

A provider reports spend against a project, but its owner is unknown. Open the source record and see what to check next.

Inspect the evidence
Sample data
Token Observe spend dashboard showing subscription bills, API invoices and metered estimates separately, using synthetic data
Review billed charges and usage estimates separately.

Your cloud workspace

It sits between routed agents and the providers you pay.

The cloud service hosts the gateway, configuration and evidence for your workspace. Requests use the separate provider inference keys you supply. Customer hosting remains available with its own operating requirements.

See the hosting model

What reaches it

Supported clients using an OpenAI-compatible, Anthropic or Gemini dialect.

OpenAI SDK · Anthropic SDK · Google Gemini clients · Claude Code · Claude Desktop · GitHub Copilot CLI (BYOK) · JetBrains AI Enterprise · Cline · Zed · Aider · Agent-to-agent callers

What changes
For those dialects, normally the base URL and the credential. No SDK swap.
Runs as
A hosted workspace in the cloud pilot, with customer hosting also available.
Cloud custody
Workspace evidence and supplied credentials are held on the hosted service.

Text by default. Inline images need an image inspector you configure; audio, PDF, remote image URLs and file references are refused.

Check your SDK and version
Your agents
Token ObserveRegistryRBACPolicyFlight recorder

campaign-writer · suspended

not registered · not governed

Routed requests → policy checks → configured providersRegistry fields and states are the product’s own; the cells are illustrative, drawn from the evaluation’s demo seed.

Runtime model compatibility

  • OpenAI
  • Anthropic
  • Google Gemini
  • OpenRouter
  • Amazon Bedrock
  • Azure OpenAI
  • Your own endpoint

Cloud setup offers OpenAI, Anthropic, xAI and Google. Use separate billing and inference credentials.

One place to look

See what AI costs, what it did and who allowed it.

The AI list shouldn’t depend on who last updated the spreadsheet.

Spend

Review the source behind each cost.

Connect a supported provider billing source. Keep reported costs by currency, with their periods and collection status. Gateway usage estimates stay separate.

Explore spend

Activity

Follow each routed request.

Follow requests through policy checks and outcomes. Turn on bounded, redacted prompt and reply capture when you need the detail. Text capture is off by default.

Explore activity

Access

Set permissions and request budgets.

Record which AI tools are approved. Give agents permissions and budgets, then govern the requests you route through the gateway.

Explore access
Capabilities and limits

Get started

Start in your browser.
Connect one source first.

Create a cloud account and workspace, then connect a supported billing source. Add agent activity when you are ready to configure a client.

  1. 01

    Create your account

    Enter your email and a password of at least 12 characters. Confirm your email using its link, then sign in.

  2. 02

    Create a workspace

    The hosted workspace starts your 30-day trial. No card and no local server to install.

  3. 03

    Connect a billing source

    OpenAI, Anthropic Claude, xAI Grok or Google Cloud Billing. Use the required reporting credential and check the source status.

Cloud pilot · no server installation
Start free 30-day trial

Billing permissions may need setup at your provider. Model calls and Google BigQuery scans can incur provider charges.

Read the cloud setup guide Instructions for agents

Try the cloud pilot

Your first 30 days are free.

The free trial
The hosted trial starts when you create a workspace. No card, server download or automatic paid conversion.
Pilot scope
This is a small pilot with shared capacity and traffic limits, without an availability SLA. Keep production-critical workloads outside it.
After the 30 days
After 30 days, workspace changes, model calls and scheduled billing collection stop. Authorized retained reads and exports remain within the shared traffic quota. Your provider charges for model calls, model-assisted search and Google BigQuery scans remain separate.
Before connecting
Cloud hosting holds your workspace data and supplied credentials. Review the security details and use only data approved for this pilot.
Trial scope and pricing

Review how your data is handled.

Check cloud custody, security controls and pilot limits before you connect your systems.

Read the security details

Choose what your first trial will check.

Start with a billing source, then add routed agent activity when you need it.

Which billing sources can I connect?

Connect OpenAI organisation costs, Anthropic organisation cost reports, xAI team API usage or a configured Google Cloud Billing BigQuery export. Each needs its reporting credential and permissions. Coverage depends on the provider account and plan.

How are costs and estimates kept separate?

Each source keeps its currency, reporting period and collection status. Provider-reported costs, imported invoices and gateway estimates remain separate, so overlapping measurements are never added together. An unconnected or failed source remains unknown.

How do I see agent activity?

Configure a supported client to route requests through your workspace with a registered agent and a separate inference key. Capture starts with metadata; bounded, redacted prompt and reply capture is opt-in. Billing connections do not recover historical conversations.

What happens after the free 30 days?

Workspace changes, model calls and scheduled billing collection stop. Authorised retained reads and exports remain subject to the pilot’s shared traffic quota. There is no automatic paid conversion. Provider model calls, model-assisted search and Google BigQuery scans can incur separate charges during the trial.

Your next step

Try it free for 30 days.

Create a cloud account and workspace. No card or server installation. Model calls and Google BigQuery scans can incur separate provider charges.

Start free 30-day trialCloud pilotExplore a sample workspace

Talk it through

Talk through your AI spend and agent setup.

Meet James Rooney, Token Observe’s founder, to choose a first evaluation. James built the product after a decade leading product and delivery in regulated organisations.

Loading the booking calendar…

No time that suits? Email hello@tokenobserve.com. Calendar not loading? Open it on cal.com.

The calendar loads Cal.com as you approach this section. Booking details go to Cal.com and James Rooney. Privacy notice.