Alerts
Push eval regressions and security events straight to Slack. Alerts are configured in the dashboard — no SDK code or redeploy required — and fire off the same evaluation and security scans already running on your traces.
Open Dashboard → Alerts (right under Overview). Everything is keyed to a single Slack Incoming Webhook; Fluiq posts to that channel server-side, so there is no bot to install and your webhook URL never reaches the browser of anyone but you.
Setup
Create a Slack Incoming Webhook
https://hooks.slack.com/services/… URL.Paste it into Fluiq
Turn on the alerts you want
Eval alerts
Team+ requiredFired from the LLM-as-judge scores produced by fluiq.eval() and auto-evaluation. Two independent triggers:
score_below
Alert the moment any watched metric scores below this value (0–1). Watched metrics: faithfulness, answer_relevancy, context_precision, hallucination, toxicity. Leave the metric list empty to watch all of them.
failure_rate_above
Alert when the share of failing evals for a metric exceeds this percentage over a rolling window of recent evaluations. Debounced — at most one failure-rate alert per metric every 10 minutes — so a sustained regression pings you once, not on every trace.
Security alerts
Growth+ requiredFired from the post-call scan behind fluiq.secure() whenever a prompt or response is flagged or blocked.
alert_on
Which risk levels page you — any combination of low, medium, high.
categories
Restrict alerts to specific attack types — prompt_injection, jailbreak, pii_detected, secrets_detected, indirect_injection. Leave empty to alert on every detected category.
blocked_only
When on, only hard blocks page you; warn-only detections are skipped. Turn it off to be notified on every flagged event.
Delivery
digest
realtime sends one Slack message per event (default). hourly and daily batch events into a single digest so a noisy window doesn't flood your channel.
How it works & cost
Alerts read the eval and security results Fluiq already computes for your traces — there is no extra scan and no added latency on your LLM calls. Delivery uses Slack Incoming Webhooks, which are free. Eval alerts are included on Team and above; security alerts on Growth and above. There is no per-alert or metered charge.
Alert delivery fails open: a webhook outage, a malformed payload, or a plan change never interrupts trace processing — a missed alert is logged and dropped, never retried into a backlog.