LLM observability icons
Traces, spans, tokens, latency, cost, evals and drift — the icons a monitoring screen for language models needs.
tracespan-tracetokentoken-countlatencylatency-p99costcost-alertcost-explorerthroughputerror-budgetdriftevaluationeval-suitedegradationguardrailcontent-filterhallucinationchart-linechart-barchart-areachart-splinemetricalertincidentlogdashboardmonitor-checkmonitor-offon-timetime-alertbudgetquotarate-limitcache-hitcache-miss
Watching a model in production means new nouns on old charts: tokens per second, cost per call, latency at p99, a prompt that drifted, an eval that regressed, a guardrail that fired. This collection pairs the AI vocabulary with the analytics and devops icons drawn on the same grid, so a dashboard is one set from the sparkline to the alert.
import { Trace, Token, Latency, Drift } from "@iconmind/react";The 36 icons, described
- Tracetrace
- Spanspan-trace
- Tokentoken
- Token counttoken-count
- Latencylatency
- p99 latencylatency-p99
- Costcost
- Spend alertcost-alert
- Cost explorercost-explorer
- Throughputthroughput
- Error budgeterror-budget
- Driftdrift
- Evaluationevaluation
- Eval suiteeval-suite
- Degradationdegradation
- Guardrailguardrail
- Content filtercontent-filter
- Hallucinationhallucination
- Line chartchart-line
- Bar chartchart-bar
- Area chartchart-area
- Chart splinechart-spline
- Metricmetric
- Alertalert
- Incidentincident
- Loglog
- Dashboarddashboard
- Monitor checkmonitor-check
- Monitor offmonitor-off
- On timeon-time
- Time alerttime-alert
- Budgetbudget
- Quotaquota
- Rate limitrate-limit
- Cache hitcache-hit
- Cache misscache-miss
Other collections
Every llm observability icon, free to ship
36 icons drawn on one grid, in outline and duotone at three weights, with the code for nine frameworks. MIT — commercial use, no attribution, no seat count.