Multi-agent DAG orchestration, purpose-built for enterprise engineering teams Learn more →

Cost meters

Last updated 2026-05-21·7 min read

The cost meter tracks real-time spending for every run. Autonomous systems can run up a bill quickly; the meter makes that spend visible and bounded. This guide covers the breakdown, cost caps, and showback.

What the meter tracks

Cost accumulates live as a run executes. The meter is visible in the run view, in the OpsBar at the bottom of the dashboard, and via the API field cost_usd on any run. It covers LLM token usage across every agent and the metered cost of cloud service calls made through MCPs.

Cost breakdown

Every run's cost decomposes three ways:

DimensionDescription
by_agentSpend attributed to each agent in the run, which agent cost the most.
by_toolSpend per MCP tool, surfaces an expensive integration call.
by_modelLLM token spend per model, input and output tokens priced separately.

Pre-commit cost caps

A run can carry a cost ceiling. The ceiling is enforced pre-commit: before the orchestrator commits a tool call that would push the run over the cap, it halts the run at a gate instead. The expensive call never executes, the run stops short of the budget rather than overshooting it and alerting you afterward.

bash
# Trigger a run with a hard cost ceiling
cendriix run ship-from-ticket \
  --ticket JIRA-3421 --max-cost 25.00
Caps as policyA cost cap can also be a workspace-wide guardrail rather than a per-run flag. A policy rule matching run_cost_usd applies the ceiling to every run automatically, see Policies & guardrails.

Estimating before you run

A dry run plans the work without executing it and prints the planned agent chain alongside an estimated cost. Use it to size a run before committing real spend.

bash
cendriix run ship-from-ticket --ticket JIRA-3421 --dry-run

# => planned: orchestrator -> dev -> qa -> sre
# => estimated cost: $0.18 - $0.42

Showback & chargeback

Cost data is retained alongside the audit log and exports for internal showback and chargeback. Tag runs with team or cost-centre metadata at trigger time, then export cost grouped by tag to attribute spend back to the team that drove it.

bash
# Tag a run for chargeback
cendriix run ship-from-ticket --ticket JIRA-3421 \
  --meta team=payments --meta cost_center=eng-42

# Export cost grouped by a metadata tag
cendriix cost export --group-by team --since 2026-05-01

How costs are priced

The meter uses published list pricing from each LLM provider and cloud service. Token counts and API call counts are exact; the dollar figure is list price multiplied by usage. Enterprise accounts with negotiated volume discounts can configure custom pricing so the meter reflects their real contracted rates.

Next steps