Service Levels
How we run maintenance and handle incidents — and what isn't covered.
Effective 2026-08-06
This is a translation provided for convenience. The Chinese version of this document governs; where the two differ, the Chinese text controls.
1. What this document does and doesn't do
This document describes how we operate the hosted service and how we handle incidents. It is not an SLA with a committed uptime figure.
For an uptime percentage, response deadlines, and service credits written into a contract, take the Enterprise plan and sign a separate agreement with us. We will not put a number on the standard plans that we cannot stand behind — we would rather this page read plainly than explain the gap afterwards.
2. Maintenance windows
Planned maintenance runs between 00:00 and 06:00 China Standard Time, announced by email at least 48 hours ahead, and usually completes within minutes.
Emergency maintenance for security patches may not meet that notice period. We will notify you beforehand where we can, and publish a write-up afterwards.
3. Incident handling
We health-check every instance. When something breaks, the order is: establish the blast radius, notify affected customers, restore service, publish a post-incident note.
Support response targets by plan:
- Starter: email support, first response within one working day;
- Team: ticket support, first response within four hours (working days, 09:00–18:00);
- Enterprise: per contract, which can include 24/7 and a named contact.
These are response targets, not resolution deadlines.
4. Backup and restore
Hosted instances are backed up daily: 7 days of retention on Starter, 30 on Team. Backups exist for our disaster recovery.
Contact support if you need a restore. Restoring overwrites current data, and we will confirm with you before doing it.
Our backups are not a substitute for yours: your code and repositories live on your own machines and backing those up is your responsibility.
5. Not covered
- failures of the machines your nodes run on, your local network, or your uplink;
- outages, rate limits, or policy changes at Claude Code, Codex, or the model providers behind them;
- failures of optional external services you configured yourself (LLM agent, Telegram / Lark, Web Push, a tunnel);
- force majeure, infrastructure-provider failure, and government action;
- data corruption or misconfiguration resulting from your own actions.