Stacklet Token Custodian

Get More Value from Every AI Token

The control plane for AI token attribution and usage governance

Book a demo

Governing our AI spend so it drives value, not waste, is exactly what we’re after, and most tools either stop at reporting or weren’t built for FinOps and governance workflows. Token Custodian does both, and it works right inside the tools our teams already use, like Claude Code and Slack, so governance doesn’t slow anyone down.

Lindbergh MatillanoDirector of Cloud & AI Optimization, Avalara

Account for Every Token with Full Attribution

Trace every token and agent run to the team, project, app, and cost center behind it, then enforce policy on it. One view across multiple providers, so spend answers to value.

Monthly token spend

Token attribution

  • OpenAI
  • Anthropic
  • AWS Bedrock
Team
Platform
Project
Support AI
App
Assistant
Tokens
1.24M
Attributed

Policy controls

Monthly budget · Engineering$8,200 / $10,000
  1. 50%Alert the VP
  2. 70%Notify the FinOps team
  3. 80%Route to efficient models
  4. 100%Approve exceptions

No hard stop. Work keeps moving.

  • Model routingThe right model for each job
    Routed
  • Runaway agentLimit: 50 tool calls per run
    Paused

Amplify AI Value with Tailored, Action-Driven Policy

Set budgets and usage policies by team, project, or environment, from model routing to spend caps. At a threshold, Token Custodian acts in real time through multi-step workflows, not just blocks, so teams keep moving and spend pays off.

Trust Governance Proven Across Enterprise Cloud

Built by the team behind Cloud Custodian, the policy engine enterprises have trusted for over ten years. The same policy-driven automation you trust for cloud, now acting on AI tokens in real time.

The Cloud Custodian approach

Cloud Custodian logoCloud Custodian10+ years in production
  • One policy languageEvery cloud, now AI
  • Built to actPolicies that take action
  • Multi-step workflowsNotify, remediate, approve
  • Real-time detectionWithin existing workflows

Key features

Key use cases

Token Attribution

Trace every token to a team, project, app, and cost center across every provider. Showback is ready for finance.

Model Routing

Send each request to the right model for the job. Routine work runs on efficient models, and premium models are saved for work that needs them.

Budget Enforcement

Set budgets by team, project, and cost center. At a threshold, route, alert, or request approval. Cap only when you need to.

Runaway Agent Containment

Set limits on agent steps, run time, and context growth. When an agent crosses one, Token Custodian takes action: pause the run, alert the owner, or kick off a workflow.

Tokens are Easy to Spend But Hard to Spend Well

See Token Custodian attribute, govern, and act on real token spend.

Take Control
Token Custodian overview: $304,280 saved, $3,412,620 spent, budgets at risk, policy actions, exceptions, attribution, and daily spend by provider

Frequently asked questions

What is Token Custodian?

A control plane for AI token usage. It attributes every token to a team, project, app, and cost center, then enforces budget and usage policy on it across every provider.

Which AI providers does it support?

Token Custodian is vendor-neutral. It governs major model providers like Anthropic and OpenAI, and leading cloud AI services like AWS Bedrock, Google Vertex, and Azure AI, all from one place.

How is Token Custodian different from an AI gateway?

A gateway manages traffic, routing and limiting per key. Token Custodian adds granular attribution and a policy layer for usage governance, tracing every token to a team or project and acting on it, like shifting to a cheaper model or routing an exception to the budget owner.

Will it block my developers when a budget runs out?

Only if you want it to. Policies escalate step by step: alert, add friction, route to efficient models, or request an approval. Capping is the last step, not the first.

How does it relate to Cloud Custodian?

Token Custodian comes from the same team behind Cloud Custodian, and applies the same proven approach: one policy language, and policies that take action in real time through multi-step workflows. Now pointed at AI tokens.

Can platform teams manage policy as code?

Yes, Token Custodian does both. FinOps can set budgets and rules in the UI, and platform teams can write complex policies as code in Git, keyed off your project metadata.