Limits
Normative for every host: dispatch time, guest memory, and payload sizes are bounded, and crossing a bound is an error the caller sees — never a truncation, never a hang. The concrete values are host operations, not something a realm can declare or rely on. The reference host’s defaults:
| Setting | Default | Behaviour at the limit |
|---|---|---|
assistant.wasm.enabled |
true |
false rejects every dispatch with a clear error — the kill switch. Declarative content is unaffected. |
assistant.wasm.dispatch-timeout |
30s | The dispatch is interrupted and returns an error. Work longer than the budget cannot run on this host. |
assistant.wasm.max-memory-pages |
1024 (64 MiB) | A module declaring more is rejected with a clear message. |
| guest I/O | 1 MiB per payload | Each of the function’s args, its result, and every host-call request/response is bounded — at 1 MiB or the module’s memory cap, whichever is smaller. Oversized payloads error rather than truncate. |
In the reference host, every dispatch is observable: it records the bundle, function, elapsed time, error if any, and how many gateway calls the guest made.

