Resources
Support
How to reach us, what to send so the first reply is useful, and what response times to expect on each plan.
Channels
| Channel | Best for | Availability |
|---|---|---|
| In-app chat | Account, billing, and quick how-to questions | Business hours, all plans |
| Email support | Anything needing logs, screenshots, or a thread | 24/7 intake, all plans |
| Discord | Peer help and informal debugging | Always on, community-run |
| Priority channel | Production incidents | 24/7, Scale and Enterprise |
| Named contact | Architecture reviews and launch planning | Enterprise |
Response targets
| Severity | Definition | First response | Update cadence |
|---|---|---|---|
| S1 — Critical | Production calls failing or agents unreachable | 30 minutes (Scale/Enterprise), 4 hours (Growth) | Hourly until mitigated |
| S2 — Major | Significant degradation with a workaround | 4 business hours | Daily |
| S3 — Minor | Isolated bug, no customer impact | 1 business day | As progress occurs |
| S4 — Question | How-to, guidance, feature request | 2 business days | On resolution |
What to include
A ticket with these five things is usually solved in one exchange. Without them, the first reply is a request for them.
- Call or session ID — From Data → Call Logs. This is the single most useful item and it unlocks the full trace on our side.
- Timestamp with timezone — Or a window. "This morning" spans four hours of logs.
- Agent and version — Which agent, which published version, which environment.
- Expected versus actual — One sentence each. What should the agent have done, and what did it do?
- Frequency — Once, intermittently, or every call. This changes the investigation entirely.
Before you file
- 1
Check the status page
Platform incidents are posted there first, usually before tickets arrive.
- 2
Open the call trace
Observability → Logging shows tool latency, model errors, and webhook failures per call. Most issues are visible there.
- 3
Reproduce in Testing Lab
A scenario that reproduces the problem turns a vague report into a fixable bug — and becomes your regression test afterwards.
Escalation
If an S1 has not had a first response within its target, reply to the ticket with ESCALATE in the first line. That routes it to the on-call engineer directly. Enterprise workspaces can also page through their named contact's channel.
Enterprise services
- Launch review — We go through prompts, failover, consent, and monitoring before your first production number goes live.
- Quarterly performance review — Containment, transfer rate, and cost per outcome, with concrete changes to try.
- Custom SLA — Uptime commitments and credits documented in your agreement.