QP
Spendagent-as-actor · grounded in model-call tokens
Agent Governance · Spend
What agents cost — estimated from the tokens in their model calls.
How this is estimated
Estimated from tokens on model-class actions. We don't see provider invoices — this is a usage-based estimate, not billed spend.
Total spend (est.)
$29.3k
59% of $50k budget
Model tokens
1.04B tok
across all model-class actions
Top agent
data-pipeline
$11.2k · 410M tok
Ungoverned spend
$0.9k
marketing-gpt · can't be charged back
Spend by agent
attributed via sourceExe (process)AgentOwnerGoverned?Model tokensSpendTop model
data-pipeline/app/pipeline.pyDatagoverned410M tok$11.2kapi.anthropic.com
cursor/Applications/Cursor.app/Contents/MacOS/CursorEngineeringgoverned320M tok$8.7kapi.openai.com
claude-code/usr/local/bin/claudePlatformgoverned210M tok$6.1kapi.anthropic.com
support-copilot/app/support_agentSupportgoverned63M tok$2.4kgenerativelanguage.googleapis.com ⚠
marketing-gpt/tmp/agent.jsno owner — unchargeablediscovered34M tok$0.9kapi.openai.com
Spend by model provider
model + unapproved-class actions, by endpointIdProvider · endpointIdModel tokensAgents using it
generativelanguage.googleapis.com ⚠ unapproved provider41M toksupport-copilot
104.18.32.7 ⚠ unapproved provider— no tokensmarketing-gpt
api.unknown-llm.example ⚠ unapproved provider— no tokensmarketing-gpt
api.anthropic.com 760M tokdata-pipeline, claude-code, cursor
api.openai.com 236M tokmarketing-gpt, support-copilot, cursor
Spend follows model tokens, attributed to the agent (process) that made the call — the same attribution the registry uses.
Fields: tokens (model-class actions) · endpointId · class · sourceExe (attribution)