# Google Perfetto v59.1 adds persistent Android traces, AI agent CLI mode, and new profiling views

> **Key takeaway:** Google's open-source Perfetto tracing system now survives Android reboots, integrates with AI coding agents, and offers call tree and function-level alternatives to flamegraphs for performance analysis.

**Published:** 2026-10-27T02:09:05+00:00 
**Source:** Google Perfetto System Profiling 
**Category:** bigtech-foss 
**Canonical URL:** https://fosswire.org/news/google-perfetto-v591-adds-persistent-android-traces-ai-agent.html 

## Summary
Google has released Perfetto v59.1, the first stable version in the v59 series after a build failure blocked v59.0 on Linux and Android builders. The release introduces persistent tracing across device reboots for Android, letting traces recover and upload automatically via a new --upload-after-reboot flag. The UI gains two alternative profile visualizations—Call Tree and Functions views—alongside a time-based Callstack flamechart that shows how thread behavior evolves rather than just aggregating samples.

The Trace Processor command-line interface receives significant ergonomic improvements including rustc-style SQL error diagnostics, markdown table output, and an agent mode that auto-detects AI coding tools via environment variables like CLAUDECODE and AI_AGENT. A new linux.smaps data source enables memory mapping capture in traced_perf, and symbolization now works without local binaries through debuginfod server integration by build ID.

## What changes in Trace Processor and profiling UI
- **Persistent Android traces:** TraceConfig.persist_trace_across_reboots preserves trace files through reboots, with recovery metadata logged in trace_recovery_reason and upload handled by perfetto --upload-after-reboot.
- **New profiling views:** Every flamegraph panel now includes Call Tree (expandable top-down/bottom-up frame tables with total and self cost) and Functions (flat per-function cost aggregation without double-counting recursion) tabs, with shared filters and no query re-execution on tab switches.
- **Callstack flamechart track:** Threads with callstack samples get a time-series visualization beneath the aggregate Callstacks track, showing per-sample stacks along the timeline with color-coding by function or library mapping.
- **Trace Processor CLI enhancements:** SQL errors use rustc-style diagnostics with source:line:col locations and carets; --format markdown outputs configurable compact tables; agent mode auto-enables markdown output and single-line JSON errors when CLAUDECODE or AI_AGENT environment variables are detected with non-terminal stdout.
- **Symbolization without local binaries:** The bundle command gains --debuginfod to fetch debug files from debuginfod servers by build ID, with local-first lookup, configurable cache paths, deduplicated cross-process mappings, and support for samply/Firefox Profiler format symbolization.

## Why it matters
Perfetto's AI agent mode and markdown output lower the barrier for automated performance analysis in CI pipelines and LLM-assisted debugging workflows, while persistent Android tracing addresses a long-standing gap in mobile reliability engineering. The debuginfod-based symbolization removes a common friction point for Linux distribution packagers and developers who need to analyze traces from stripped production binaries without maintaining local debug file archives.