Skip to main content

Getting help

platform v0.9.11verified 2026-05-14

A support request is fastest to resolve when the first message already contains everything we need. This page lists what to attach for each audience so we don't have to ask.

Where to send it

Support contact details depend on your platform deployment. If you reached Delphi via a customer-operated install, contact that team first; they'll escalate to KI Kombinat if the issue isn't tenant-visible. For the hosted platform at app.delphi.ki-kombinat.com, use the contact channel on your contract.

What to include​

Always​

  • A one-sentence summary of what you tried, what you expected, and what happened instead.
  • The URL of the page or app where it happened. For Flow Designer issues, include the flow ID (visible in the URL).
  • When it happened (date, time, timezone). A 5-minute window is enough.
  • The Delphi platform version — see Version sources. For tenants, your team admin can read this from the TelWeb footer or the admin view.
  • The environment: hosted (app.delphi.ki-kombinat.com), staging, your own install, etc.

For users​

  • The browser and version (e.g. "Chrome 132 on macOS 15").
  • For audio / WebRTC issues: whether the microphone is granted, and the result of the in-page audio device test if your platform exposes one.
  • A short screen recording when the symptom is visual or interactive; a screenshot when it's a single frame.
  • The conversation ID if the issue is about a specific conversation (visible under the app's conversation history).

For team admins​

  • The app name and flow name affected.
  • For tool / variable issues: the tool name and the most recent conversation ID that demonstrates the problem.
  • For API issues: the API key prefix (never the full secret), the endpoint, the HTTP status returned, and a request/response sample with secrets redacted.
  • For billing issues: the invoice or session ID from Stripe, not the full card data.

For platform operators​

  • The service (web, api, voice, telpro, database, media, ops, squid, signoz).
  • The host identifier and the container name (voiceai-*).
  • The release tag (ECR_TAG) currently deployed.
  • A relevant log excerpt (10–50 lines, including the ERROR line and a few surrounding lines for context).
  • A SigNoz trace ID when the issue spans services — see SigNoz monitoring for finding one.
  • Whether the issue affects all calls / all users or a subset. If a subset, the smallest reproducer you have.

A workable template​

Summary:        AI doesn't respond on inbound calls to +49 ... since 09:30 CET.
Platform:       v0.9.11, hosted at customer-install.example
Environment:    production
Service:        voice (voiceai-telphi)
Release tag:    v0.9.11
Host:           prod-voice-02
Trace ID:       (from SigNoz)
Logs:
  <paste 20 lines around the ERROR>
What I tried:   restarted voiceai-telphi, checked HTTP_PROXY, Squid is healthy.
Impact:         all inbound calls; outbound webrtc test calls work.

Before you file: the 90-second checklist​

Most tickets fall into one of these buckets. Checking them first often shortcuts the whole exchange.

  1. Is there a banner or status page incident? Many platform installs ship a status page.
  2. Is the feature behind a flag? Look for a FeatureFlag callout near the feature in these docs and ask your platform admin if unsure.
  3. Did anything change recently? A recent release, a config change in vars.yaml, a rotated secret, a new firewall rule? See Release notes.
  4. Is it reproducible in another account / browser / network? Helps separate tenant vs platform issues.
  5. Is SigNoz telling you anything? For operators, the observability dashboards often pinpoint the failing service before logs do.

How requests are triaged​

  • Sev 1 — Platform down, all calls failing, billing/payment outage. Operator escalation path applies; SLA depends on your contract.
  • Sev 2 — A major flow is broken (e.g. all calls into a single endpoint fail, publish breaks for a tenant). Typically responded to within hours.
  • Sev 3 — Single-user or visual issue, workaround exists. Triaged within the next business day.
  • Sev 4 — Feature requests and documentation gaps. Tracked but not on a clock.

If your issue is Sev 1 or Sev 2, say so explicitly in the summary — the triage will not promote it for you.