feat: In-CLI Performance Monitoring Dashboard (/perf command)

[Shivangisharma4]

Summary

Add a /perf slash command that surfaces Gemini CLI's existing telemetry data as a real-time, in-terminal performance dashboard. This gives developers immediate visibility into startup timing, memory consumption, tool execution profiling, model API latency, and session statistics, without leaving the CLI.

Motivation

Gemini CLI ships a rich but hidden telemetry layer - OpenTelemetry-based metrics, a startup profiler, a memory monitor, activity tracking, and regression detection are all instrumented but never surfaced to the user. The existing /stats command covers token/quota info, but there is no way for users to:

• See how long startup took and which phases are slow
• Monitor memory consumption during a session
• Profile which tools are slowest or called most frequently
• Track API latency per model across requests
• Detect performance regressions between versions
• Export reports for CI integration or bug reports

Existing Infrastructure

Component	Location	What It Does
StartupProfiler	packages/core/src/telemetry/startupProfiler.ts	Phase-based startup timing with performance.mark/measure
MemoryMonitor	packages/core/src/telemetry/memory-monitor.ts	Continuous heap/RSS monitoring, high-water marks, thresholds
ActivityMonitor	packages/core/src/telemetry/activity-monitor.ts	Event buffering with listener pattern
metrics.ts	packages/core/src/telemetry/metrics.ts	50+ OpenTelemetry counters/histograms
/stats command	packages/cli/src/ui/commands/statsCommand.ts	Existing pattern for session stats via SlashCommand
SessionContext	packages/cli/src/ui/contexts/SessionContext.tsx	Real-time metrics state with event subscription

Proposed Design

/perf Slash Command with Sub-Commands

/perf                 → Overview dashboard (all sections)
/perf startup         → Startup time breakdown by phase
/perf memory          → Current memory usage + high-water marks
/perf tools           → Tool execution timing & frequency table
/perf api             → API latency per model + error rates
/perf session         → Session lifetime statistics
/perf export [format] → Export report as JSON/Markdown

Core Data Aggregation Service (PerformanceDataService)

• Aggregates data from StartupProfiler, MemoryMonitor, ActivityMonitor, and session metrics
• Exposes a unified getPerformanceSnapshot() API
• Computes derived metrics: avg latencies, success rates, memory usage percentages

Ink-Based Dashboard Rendering

• Color-coded startup phase breakdown (green/yellow/red thresholds)
• Memory usage bar with heap percentage and RSS tracking
• Tool profiler table sorted by total execution time
• Per-model API metrics with error rate indicators

Implementation Phases

1. Data Layer - PerformanceDataService aggregating existing telemetry sources
2. /perf Command & UI - Slash command + Ink components following statsCommand.ts pattern
3. Export & CI - JSON/Markdown exporters, regression detection with configurable thresholds
4. DevTools Integration (stretch) - Stream metrics to DevTools via existing WebSocket infrastructure

Expected Outcomes

• /perf command accessible in-CLI with sub-commands
• Startup time breakdown with optimization suggestions
• Memory usage monitoring with high-consumption warnings
• Tool execution timing and frequency statistics
• Model API latency tracking per request
• Performance regression detection for CI integration
• Export capability for performance reports

Related

• Telemetry sources: packages/core/src/telemetry/
• DevTools: packages/devtools/

[gemini-cli]

There is a problem with the Gemini CLI issue triaging. Please check the action logs for details.

[Shivangisharma4]

/assign

[gemini-cli]

👋 @Shivangisharma4, thanks for your interest in this issue! We're reserving self-assignment for issues that have been marked with the help wanted label. Feel free to check out our list of issues that need attention.