Cost Controls

See every dollar of AI spend,
and cap it.

Track tokens and cost across every team, user and agent, set budgets and token-aware limits, and stop runaway spend in real time.

  1. RequestFrom a user, app or agent
  2. Fabric
    1. IdentifyTeam, user and agent
    2. LimitsBudgets and token caps
  3. ProviderTokens and cost recorded
Every request is attributed before it is sent, limits apply before the provider is called, and cost is recorded on the same trace.
Team, user, agent
spend attribution
Real time
limit enforcement
Every call
tokens and cost counted

The problem

AI spend grows one team and one agent at a time. Provider invoices arrive as a single number, so nobody can say which team, user or agent drove it, and the first sign of a runaway agent loop is the bill. Limits set in each application drift, and they never cover the agents nobody registered.

Solution

Every model call goes through the Fabric data plane with a verified identity attached, so Fabric can count tokens and cost on the way through and attribute them to the team, user and agent behind the request. Analytics breaks spend down along those lines and by route, model and namespace. Token-aware limits stop a request before it reaches the provider once a budget is used up, and the killswitch cuts off a scope immediately. To lower the cost of each request as well, pair it with Adaptive Model Routing, Lightyear and MCP Tool Search.

What you get

Spend by team, user and agent

Tokens and cost attributed to whoever made each request.

Budgets and token limits

Cap requests and tokens per team, route or consumer, enforced in the gateway.

Cost in analytics and traces

Break spend down by route, model and namespace, down to a single trace.

Killswitch

Block traffic for a tenant, service, route or tool immediately.

Per-namespace views

Each team sees and manages its own spend.

Frequently asked questions

  • How does Fabric attribute AI spend to teams, users and agents?

    Every request reaches the Fabric data plane with a verified identity attached. Fabric counts the tokens and cost of each call on the way through and records them against the team, user and agent behind it, so analytics can break spend down along those lines.

  • What happens when a team reaches its limit?

    Token-aware limits are enforced in the gateway: once a limit is used up, the request is rejected before it reaches the model provider, so it is never billed.

  • Can I stop a runaway agent immediately?

    Yes. The killswitch blocks traffic for a tenant, service, route or MCP tool scope at once, without a deploy.

  • Does Fabric charge for LLM tokens?

    No Fabric meter applies to LLM provider traffic on the Free plan. You pay your model provider at their list price, and Fabric adds routing, policy, limits and cost visibility on top.

  • How do cost controls relate to reducing AI cost?

    Cost controls show and cap what is spent. To lower the cost of each request as well, use Adaptive Model Routing, Lightyear and MCP Tool Search on the same routes.

Bring every AI interaction under control

Connect your first model, MCP server or internal API and apply consistent governance from one gateway.