Mezmo Insights
Know what changed.Give your agents context.
Put anomaly detection on autopilot. Mezmo surfaces anomalies with zero setup and turns logs, metrics, and traces into focused evidence your agents can use. Start investigating with context already in hand.
Automatic anomaly detection. Zero setup. Context ready for your agents.
Logs → changed patterns. Repeated messages become evidence your agent can inspect.
Metrics → expected behavior. Cadence, variability, and seasonality give each series context.
Traces → focused paths. Service calls and failed spans show where to look next.
Log Insights
The shape of your logs.The changes that matter.
See errors, warnings, and event volume against the previous period. Find the applications and hosts that shifted, then inspect the log patterns behind the change. No query needed to get started.
Example Log Insights view with fictional data: errors, warnings, and events from 09:00 to 15:00, compared with the period before. Errors rose from 172 to 1,528. The checkout app accounts for most of the change (1,248 errors, up from 96), and the changed pattern behind it is “Connection to <*> timed out after <N> ms”.
Log Insights
Updated just nowAccount timeline compared with baseline
now beforeApps whose errors changed most
Hosts that changed most
Error patterns that changed
Explore tabs, time ranges, a pattern, and “View in context”. All values are fictional.
See activity against a baseline
Compare a time window with the period before or the same time yesterday.
Find the changed pattern
Variable values are normalized so repeated messages become a pattern you can track.
Jump to the actual line
Open a recent log entry in context. Share a view with its time range and comparison intact.
Metric Insights
Know how a metric behaves.Then inspect the change.
Anomalies surface automatically, with zero setup. Mezmo learns each series’ cadence, variability, and seasonality so you can go straight to the flagged windows and inspect the values behind them.
Example Metric Insights view with fictional data: 1,248 metric series profiled by emission pattern (continuous, scheduled, irregular), variability (stable, variable, spiky), and seasonality (daily, weekly, none), with 3 anomalies flagged in the past day for checkout, batch-worker, and worker.
Metric Insights
Profiles and detections across your metric seriesVariability by emission pattern
Select a cell to explore that emission pattern.
fewer seriesmore series
Seasonality
Recurring patterns in series behavior.
Available to your agent through get_metric_analysis_overview
Anomalies by app, past day
Explore the catalog, profiles, and flagged windows. All names and values are fictional.
Give each series a profile
Continuous or scheduled. Stable or spiky. Daily or weekly patterns add context before you interpret a change.
Go from a finding to its values
Explore flagged windows and the underlying five-minute data. Keep the finding separate from a conclusion.
Give your agent the same context
Query profiles and anomalies through MCP, then retrieve a focused window of series values.
Trace Insights
Follow the call.See what happened.
Explore services and their dependencies in a map built from OpenTelemetry traces. Inspect latency on each connection, then open recorded failed traces to see the spans involved.
Example Trace Insights view with fictional data: a service map built from traces, where frontend calls checkout, and checkout calls cart and orders. Checkout is selected, with a p95 latency of 1.12 seconds over 6,280 calls, and links to its recorded failed traces and their spans.
Trace Insights
Latest data 15:00Map colors use fixed p95 thresholds, not a comparison with baseline. All values are fictional.
See your service relationships
Explore incoming and outgoing calls without drawing a service map by hand.
Go from service to failed span
Inspect a failed trace and its failure details, rather than stopping at a slow connection.
Ask about changes through MCP
Agents can query service calls that became slower than their baseline, separately from the map’s latency thresholds.
What your agents get
A head start.And a path to the evidence.
Give your agent precomputed findings through Mezmo MCP, so it can begin with patterns and behavior instead of assembling every answer from raw telemetry.
Less repetition to work throughGet a normalized log pattern, its change against baseline, and the time window to inspect.
A sense of what is expectedLearn whether a metric is continuous, scheduled, spiky, or seasonal before examining its values.
A narrower next stepFind service calls worth a closer look, then retrieve a recorded failed trace and its spans.
Use the signal-specific tools from your MCP-connected agent. Your agent remains in charge of the investigation.
Example Mezmo MCP exchange, shown for log patterns, metric behavior, and service calls. For logs: the agent asks what changed in checkout’s logs, the find_log_patterns tool returns the changed pattern with its counts (1,248 events against 96 before) and time window, and the agent follows up with get_log_pattern_series for that pattern’s series.
What changed in checkout’s logs?
find_log_patternsConnection to <*> timed out after <N> ms- app
- checkout
- events
- 1,248 vs 96 before
- window
- 09:00 to 15:00
get_log_pattern_seriesPass the returned template_id and a time range. Raw examples require include_exemplars: true.
Your agent chooses the next query and follows the evidence. Mezmo MCP documentation
You found the change.What comes next?
Insights gives your agent the evidence. Memory adds experience. Governance sets the boundaries. Ledger shows how the evidence was used.
Has this happened before?
Bring the current finding into context with prior incidents, the actions taken, and their verified outcomes. Reuse lessons your team has reviewed.
Close the loop.
Your agent understands its environment. Governance helps it act within your policy.
Explore GovernanceHow did the agent use the insight?
Follow the finding into the agent’s decisions, tool calls, and observed outcomes. See what evidence it had and what happened next, in one linked record.
From the original finding to the recorded outcome, keep the investigation connected.
Better evidence for people.Better context for agents.
Logs, metrics, and traces. Ready for the next question.
