Cereus Insights for Cacti has received a significant update. The release introduces a new Threshold Suggestions engine, expands LLM support to three providers, and tightens how the system handles inverted metrics like free memory and available disk space. Here is what is new.
Threshold Suggestions (Professional+)
The most visible addition in this release is the Threshold Suggestions page. Many Cacti installations have hundreds of datasources with no Thold thresholds configured — not because operators don’t want alerting, but because setting thresholds manually across a large environment is time-consuming and requires knowing what “normal” looks like for each metric.
Cereus Insights now analyses the baseline data it already computes for anomaly detection and derives a recommended thold_hi or thold_low value for every unconfigured datasource. Each suggestion comes with a confidence score: the percentage of the 168 hour-by-day-of-week time buckets that have sufficient history. A score of 100% means the system has seen the metric at every hour of every weekday for the full baseline window — the suggestion is grounded in complete data.
Direction is handled automatically. Datasource names containing “free”, “avail”, or “idle” receive a thold_low suggestion and will alert when the value drops below the threshold, not above it. All other datasources receive a thold_hi suggestion. One-click Create applies the value to Thold directly. Skip dismisses a row permanently without affecting other analysis. The Exclude DS Name control adds a datasource name to a global exclusion list that suppresses it across suggestions, anomaly detection, and forecasting simultaneously.
Enterprise-tier users also get an Explain button that sends the baseline data to the configured LLM and returns a plain-English description of why that threshold value was chosen.
Multi-Provider LLM Support (Enterprise)
The LLM alert summarization feature previously required an Anthropic API key. It now supports three providers: Anthropic (Claude), OpenAI (GPT), and Google (Gemini). A Provider dropdown in the Settings tab controls which provider is active. The “Test API Key” button validates the key against the selected provider before it is saved, so misconfigured keys are caught immediately rather than at the next alert batch.
The LLM API Key label in Settings has been updated accordingly — it is no longer labelled as a Claude-specific key.
Forecast Direction Awareness (Community+)
Capacity Forecasting now distinguishes between metrics that grow toward a ceiling and metrics that drain toward a floor. For datasources whose names contain “free”, “avail”, or “idle”, the forecast targets thold_low instead of thold_hi and only projects a saturation date when the slope is negative — when the value is actively declining. Growing free space generates a “Never / Stable” result rather than a spurious future date.
This eliminates a category of false urgency that appeared on disk-free and memory-available datasources in previous versions.
Other Changes
The global datasource exclusion list, managed from the Threshold Suggestions page, also reduces baseline computation and forecast work. Adding a noisy datasource name to the list removes it from every processing pass, which is useful for constant-value counters or metrics with frequent resets.
The Status box now shows four panels: Anomaly Baselines (detection readiness percentage, average samples per bucket), Capacity Forecasts (quality percentage, average R²), Threshold Suggestions (eligible count, high-confidence count), and AI Summaries. During an active batch pass the panels display an in-progress indicator rather than a static countdown.
Batch size defaults to 500 and forecast recomputation runs every hour.
Availability
Cereus Insights requires Cacti 1.2.x or later. Threshold Suggestions and Anomaly Detection require a Professional license. LLM Summarization and the Explain button require an Enterprise license. Capacity Forecasting is available on all tiers.
Licenses and plugin downloads are available at urban-software.com