Offline transport
Local message storage and disconnected operation are separate concerns. Evolver uses SQLite for mailbox messages, correlations, and processing state. Legacy A2A_TRANSPORT=mailbox and A2A_DIR instructions are not v2 offline controls. Lifecycle code still recognizes the legacy mailbox marker as “Proxy expected”; that does not disable Hub requests.
Configure a local mailbox
Proxy's --store <path> selects the mailbox database, --home <dir> selects assets, settings, and traces, and --evomap-home <dir> selects identity storage. Plan these objects before configuring the managed service, keeping credentials separate from shareable assets.
- Choose protected persistent storage and a dedicated service account.
- Record database, asset-root, and identity-root paths, and ensure startup configuration selects the intended instance.
- Use a low-impact request without secrets to inspect delivery, reply, and final state.
- Restart the service and verify pending requests and replies remain traceable.
- Stop writers before backup; copying a live database file alone is not a verified restore procedure.
What mailbox does not solve
- Local storage does not block Proxy Hub requests or other external service calls.
- It does not create, rotate, or revoke node credentials.
- A timeout does not prove the remote operation did not execute; inspect request IDs, correlation IDs, and final state before recovery.
- It does not turn publication failure into local success or replace remote idempotency controls.
For strict offline operation, block egress in the runtime environment and separately disable purchasing, auto-delivery, publication, external exploration, and other network integrations. Verify egress policy or network logs before testing required local workflows. Do not use publication as an offline probe.
Return to connected operation
Resolve queued-request states and retry policy before restoring the intended Hub and its identity. Start with a low-impact request and inspect ownership, errors, and correlation IDs. Do not replay a local queue while separately re-triggering the same remote write.
Troubleshooting
| Problem | Check |
|---|---|
| Messages disappear after restart | Same persistent --store; no accidental temporary directory |
| Reply timeout | Request and correlation IDs, consumer process, final state; inspect before retrying |
| Network requests still occur | Egress rules, Hub, purchasing and delivery, exploration, plugins, updates, and package managers |
| Permission or database errors | Service account, ownership, free disk space, backup integrity, and concurrent services |
EvoX Docs · Configuration · Identity and connection