# 5.8 — Rate limiting Tier 3 (atlas) (GREEN)

Phase: 5 — Chat + Rate Limiter (Surface D)
Stage: GREEN

Depends on: [5.7](5.7-api-chat.md)

- [ ] Global chat semaphore = 6 (2 of 8 GPU slots kept as operator headroom); queue depth 20 then reject `429`; per-session budget 12 msg/24h; per-request timeout 120s hard server-side; disconnect cancels upstream immediately (`req.Context()` threaded to vLLM request)

## Verify

```bash
# load test at 20 concurrent clients
# GPU sequence usage never exceeds 6, no upstream 5xx, queue drains
```