AI features
Suggested replies, ticket summarization, and AI-assisted KB search. Each is opt-in per org, audited, and quota-capped per plan.
- Where:
- Settings → AI · Agent workspace → AI panel
- Permission:
- MANAGE_SETTINGS
- Updated:
- July 2026
Overview
- Agent-assist surfaces: Suggested replies, Reply tone polish, Request summarization, AI-assisted KB search, Similar past tickets (semantic — no tokens), and Draft-a-KB-article from a resolved ticket.
- Intake automation: AI triage suggests a priority, request type, and team on new requests (suggest-only — an agent applies it, nothing is changed automatically), and sentiment/at-risk detection flags frustrated tickets in the queue.
- Portal + reporting: a self-service resolver answers employees from your published KB before they file a ticket, a weekly insights digest clusters the week's requests into themes, and natural-language reporting turns a plain-English ask into a filtered queue view.
- Every surface is opt-in per organization and enabled separately in Settings → AI. All are off until an admin turns them on.
- Token note: triage and sentiment run automatically on each new request (and sentiment re-runs on each requester reply) once enabled, so their token use scales with ticket volume rather than agent clicks — watch the usage dashboard when enabling them at high volume. Similar-tickets, insight clustering, and cached replies consume no completion tokens.
- Provider is abstracted: Anthropic, AWS Bedrock, or any OpenAI-compatible endpoint (OpenAI, Azure OpenAI, Ollama, vLLM, and similar) reached via AI_PROVIDER=openai_compatible plus AI_BASE_URL.
- Usage is logged per org / user / feature / token-count / outcome. Plan caps pause AI features when the monthly token budget is reached — no overage charges.
Step by step
- 01
Enable AI
Settings → AI. Toggle on the surfaces you want. SaaS uses Anthropic by default. Self-hosted selects a provider with the AI_PROVIDER env var.
- 02
Set provider credentials
SaaS: handled by EvenTier. Self-hosted: set the provider key (e.g. ANTHROPIC_API_KEY, or AI_API_KEY for an OpenAI-compatible endpoint, or AWS credentials for Bedrock). Restart the worker after rotating keys.
- 03
Watch quota
Settings → AI → Usage. The dashboard shows tokens by feature and projects month-end consumption. Approaching the cap, AI features pause cleanly with a banner explaining why.
- 04
Disable for an airgap deploy
Set AI_DISABLED=true in your Helm values. The provider is never initialized, the AI surfaces hide entirely, and no telemetry leaves the cluster.
Good to know
- If suggested replies seem off-tone, the KB articles you have published bias the model — refine the most-cited articles first.
- AI-assisted KB search, Similar tickets, and the portal resolver ride on the same semantic + trigram search — improving article titles and content improves all of them.
- Enabling AI features one at a time and watching the by-feature usage dashboard for a week is the cleanest way to size your token bucket before turning on the always-on intake features (triage, sentiment) at full volume.