AGT-11 · Agent cost & step-budget control¶
Budgets for agents live in policy documents, while a single runaway loop or a chatty tenant can burn a month's model spend in hours. Azure budget alerts arrive a day late, so teams need ceilings enforced in code and at the gateway, and a fast way to find who spent what.
- Who reads it: Agent engineers, platform engineers and FinOps-minded DevOps engineers
- Outcome: Hard per-run, per-user and per-tenant ceilings enforced in code
- Pack owner: Sivaraj (catalog owner role: Agent Lead)
Master catalog entry¶
- Catalog name: Agent Cost & Step-Budget Controller (P2)
- Why it matters: A runaway agent loop is the fastest way to lose a client's cloud budget
- What it removes: Hard budget ceilings enforced in code rather than in policy documents
- Stack: Python middleware, Redis counters, APIM policies
- Maturity target: Level 2 — per-run, per-user and per-tenant budgets
Playbooks¶
| ID | Story | Title | Kind | Complexity | Urgency | Status | Owner | Planned publish |
|---|---|---|---|---|---|---|---|---|
| PB-AGT-11-01 | AGT-11-S1 | Enforce per-run, per-user and per-tenant budgets with Redis counters | How-to | Medium | 4 | draft | Sandeep Velu | 2026-12-09 |
| PB-AGT-11-02 | AGT-11-S2 | Roll out or change API Management token limits per tenant | Runbook | Medium | 4 | draft | Sandeep Velu | 2026-12-01 |
| PB-AGT-11-03 | AGT-11-S3 | Investigate an agent cost spike | Runbook | Medium | 5 | draft | Sandeep Velu | 2026-11-17 |
| PB-AGT-11-04 | AGT-11-S4 | Set budget ceilings and choose what happens at the ceiling | Decision guide | Low | 3 | draft | Sandeep Velu | 2026-12-21 |