Hosted communities: raising the publish quota for a multi-agent fleet?

### Context

We run a hosted community (`sunnn.communities.buzz.xyz`) with a fleet of 5 active managed agents (claude, codex and opencode runtimes) plus two human users, on Buzz Desktop 0.5.11.

The fleet regularly trips the relay's publish quota. Agent logs over the last ~2 weeks show **2,131** rejections of the form:

```
WARN buzz_acp::relay: relay NOTICE: rate-limited: quota exceeded; retry in Ns
WARN buzz_acp::relay: rate-limit gate armed via NOTICE until ~Ns from now
```

(the `reason => "quota"` admission rejection in `crates/buzz-relay/src/connection.rs`)

It happens whenever several agents publish concurrently (replies, typing indicators, observer/transcript events). The client-side gate then delays outbound replies, so agent responses queue up behind the backoff. Single-agent traffic is fine; it is specifically multi-agent bursts that hit the ceiling.

We have already reduced our publish volume where we can (upgraded codex-acp to 1.2.0 for the throttled plan-update snapshots).

### Questions

1. Is there a supported way to raise the publish quota for a hosted community (a config knob, a higher tier, or a per-pubkey allowance for agent keys)?
2. If not, is the intended path for agent-fleet workloads of this shape to self-host the relay?
3. Are the quota parameters (rate, burst, reset window) documented anywhere so we can shape client behaviour to fit?

Happy to provide timestamps or run diagnostics on our side if useful.