Summary
Per-session budget enforcement plugin that observes and enforces lifetime limits on cumulative tokens, inference calls, and wall-clock duration at the sidecar layer — no agent code changes required. Backed by Redis for cross-pod durability with a local in-memory cache for zero-latency evaluation on the request path.
Out of scope
- HITL pause mode (webhook approval)
- OTel metrics emission
- Historical spend reporting / audit log
- Cross-harness session identity standardization
Part of epic #708
Summary
Per-session budget enforcement plugin that observes and enforces lifetime limits on cumulative tokens, inference calls, and wall-clock duration at the sidecar layer — no agent code changes required. Backed by Redis for cross-pod durability with a local in-memory cache for zero-latency evaluation on the request path.
Out of scope
Part of epic #708