Run cost-effective, always-on agents to monitor your environment: the engineering behind Grafana Assistant Watchers
Blog post from Grafana Labs
Grafana Watchers, generally available through Grafana Assistant as of ObservabilityCON 2026, are always-on monitoring agents designed to identify issues that predefined alert rules may miss while keeping AI costs low. Users describe what to monitor and select telemetry sources, after which a one-time calibration agent uses a capable model to map signals, learn normal behavior, incorporate existing dashboards and alerts, and save checks and contextual guidance. During recurring runs, conventional code performs data queries, baseline comparisons, log-pattern tracking, and deployment checks, while a single small, fast language-model call evaluates the prepared evidence and decides whether to report, notify, or launch an investigation. Grafana reports that 41 of 43 internal watchers cost under $10 monthly, with a median below $5, compared with potentially thousands of dollars for an agentic investigation on every run. Internal use reportedly uncovered more than 10 customer-affecting production bugs, though engineers remain responsible for validating root causes. Watchers support numerous Grafana and external data sources, HTTP APIs through Infinity, and notification destinations including Slack, Microsoft Teams, webhooks, Grafana Alerting, and Assistant Investigations; they are free until December 1, 2026, after which they will consume Assistant token usage.
| Trend | Post Mentions | Total Month Mentions | Posts | Companies | MoM |
|---|---|---|---|---|---|
| LLM | 6 | 1,192 | 304 | 123 | -82% |
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.