Where the agent tokens actually go.
A workflow fans out to parallel agents, so it spends more tokens, not fewer. This view shows where they go, by phase and by agent, flags redundant work, and reads back the agents' captured reasoning. Observability, not a "fan-out saves tokens" claim.
Honest token framing. A workflow (ultracode) fans out to parallel agents, so it spends more tokens than a single agent, not fewer. This view shows where the tokens go (by phase, by agent), flags redundant agent work, and trends tokens-per-task (which should fall as memory compounds). The value is observability and redundancy-hunting, not a "fan-out saves tokens" claim.
Agentic insights on Insights →
Tenant-isolated by Postgres RLS · you only ever see your own org's workflows, agent runs and reasoning. Token counts are reported as measured. Reasoning content is decrypted server-side with the per-org key, for display only.