Gemini CLI v0.69.0 maps your token usage and stops silent network hangs

1 day ago

Gemini CLI v0.69.0 adds /stats heatmap, a calendar of your daily token usage, and stops the silent multi-minute spinner when the network fails. It also stops hours of quota retries, lets long-thinking requests wait past 60 seconds, fixes out-of-memory crashes and resume errors in long sessions, and fixes piped input, /model with a name, standalone theme.json themes, PowerShell commands with parentheses and macOS sandbox folder trust.

Ask about this presentation

Chapters

  1. 0:00A calendar of your token usage
  2. 0:12Name a day and the heatmap breaks it down
  3. 0:2501: When the network fails
  4. 0:27A failing network now says so
  5. 0:39An exhausted quota stops instead of waiting hours
  6. 0:50Long-thinking requests can wait past 60 seconds
  7. 1:0102: Long sessions
  8. 1:02Long sessions stop running out of memory
  9. 1:15Resuming an interrupted session works
  10. 1:2803: Scripts and commands
  11. 1:29Headless runs get every byte and the right exit code
  12. 1:41/model takes a model name and checks it
  13. 1:53Custom colors can live in their own theme.json
  14. 2:06Gemini CLI won't start a second copy inside shell mode
  15. 2:1804: Windows, macOS and VS Code
  16. 2:20PowerShell commands with parentheses run again
  17. 2:33The macOS sandbox remembers trusted folders
  18. 2:47The VS Code companion is 80 percent smaller
  19. 3:01Should you upgrade?
Show transcript
  1. Gemini CLI now draws your token usage as a calendar. The stats heatmap command shows 28 weeks of daily usage in GitHub style, with your current streak and your longest.

    Slide text
    Gemini CLI

    Gemini CLI v0.69.0 maps your token usage and stops silent network hangs

    > /stats heatmap
    Token Usage Heatmap: 28 weeks of daily token usage with current and longest streak
    packages/cli/src/ui/components/HeatmapComponent.tsx · #30135
  2. Add a date, or yesterday, and the heatmap splits that day's tokens into input, output and cached, with each model's share. It counts local chat sessions only.

    Slide text
    > token usage 2026-11-19
    > /stats heatmap yesterday
    Day breakdown: token split, sessions, turns and tokens per model
    docs/reference/commands.md · #30135
  3. When the network fails.

    Slide text
    > when the network fails 2026-11-19
    01
    When the network fails
  4. A dropped connection could leave Gemini CLI spinning for up to twenty minutes with no error. Now each network retry shows on screen, and an unreachable host fails on the first try.

    Slide text
    > when the network fails 2026-11-19

    A failing network now says so instead of spinning

    ⠏ Trying to reach <model> (Attempt X/Y) ENETUNREACH · EHOSTUNREACH · bad TLS certificate → fails on the first attempt
    packages/cli/src/ui/hooks/useLoadingIndicator.ts · #30182
  5. When your quota ran out with a reset hours away, Gemini CLI kept retrying until the reset. Now it stops at once, and an interactive session offers the fallback prompt.

    Slide text
    > when the network fails 2026-11-19
    429, quota resets in hours: retried until it resets ✓ the request fails at once ✓ interactive sessions get the fallback prompt ✓ ACP editors see each retry
    packages/core/src/utils/retry.ts · #30187
  6. A request that thought for over a minute before answering was dropped and sent again. Now you can raise that 60-second limit, on Vertex AI and API keys too.

    Slide text
    > when the network fails 2026-11-19
    settings.json model.requestTimeoutMs environment GEMINI_API_HEADERS_TIMEOUT_MS Vertex AI · API key · Sign in with Google
    docs/reference/configuration.md · #30183
  7. Long sessions.

    Slide text
    > long sessions 2026-11-19
    02
    Long sessions
  8. Sessions that kept reading files, screenshots or PDFs could crash with a JavaScript heap out of memory error. Now older images and PDFs leave the history, and tool output is capped.

    Slide text
    > long sessions 2026-11-19

    Sessions full of files, screenshots and PDFs stop running out of memory

    FATAL ERROR: Reached heap limit Allocation failed - JavaScript heap out of memory ✓ older images and PDFs leave the history ✓ tool output capped at 64 KB per call
    packages/core/src/context/chatCompressionService.ts · #30180 #30217
  9. If you stopped Gemini CLI while the model was thinking, resuming could fail with a 400 error. That is fixed, and long session files load without freezing the terminal.

    Slide text
    > long sessions 2026-11-19

    Resuming an interrupted session no longer ends in a 400

    400 INVALID_ARGUMENT: Requests ending with a model turn are not supported. ✓ interrupted turns are repaired on resume ✓ session files load off the main thread
    packages/core/src/core/geminiChat.ts · #30186 #30153
  10. Scripts and commands.

    Slide text
    > scripts and commands 2026-11-19
    03
    Scripts and commands
  11. In headless mode, Gemini CLI dropped piped input slower than half a second. Now it waits for the whole pipe, kills its child shells on Ctrl+C, and exits with the right code.

    Slide text
    > scripts and commands 2026-11-19

    Scripts that pipe into gemini -p get every byte and the right exit code

    ✓ waits for a slow pipe instead of dropping it after 500 ms ✓ Ctrl+C kills its child shells and exits 130, SIGTERM 143 ✓ errors print a message instead of [object Object]
    packages/cli/src/utils/readStdin.ts · #30163 #30204 #30117
  12. The model command ignored a name typed after it and opened the picker. Now it switches straight to that model, rejects unknown names, and tab-completes valid ones.

    Slide text
    > scripts and commands 2026-11-19

    /model now takes a model name and checks it

    > /model flash-lite ✓ switches to it, no picker > /model <typo> Invalid model "<typo>". Run /model to view and select from available models.
    packages/cli/src/ui/commands/modelCommand.ts · #30137
  13. Custom themes can now live in their own file, theme.json, in your Gemini folder. Palette generators like pywal and matugen can rewrite it without touching your settings.

    Slide text
    > scripts and commands 2026-11-19
    ~/.gemini/theme.json { "custom-wallust": {   "type": "custom",   "background": { "primary": "#181818" } } }
    docs/cli/themes.md · #30119
  14. Typing gemini in shell mode started a second Gemini CLI inside the shell pane, and the two fought over the screen. Now that is blocked, and headless prompts still run.

    Slide text
    > scripts and commands 2026-11-19

    Gemini CLI won’t start a second copy of itself in shell mode

    ! gemini → a second Gemini CLI fighting the first for the screen ✓ blocked, pointing you to /model and /settings ✓ gemini -p, --version and gemini mcp still run
    packages/core/src/utils/shell-utils.ts · #30185
  15. Windows, macOS and VS Code.

    Slide text
    > windows, macos and vs code 2026-11-19
    04
    Windows, macOS and VS Code
  16. For several releases, Gemini CLI blocked ordinary PowerShell commands with parentheses as command substitution. They run again, and a substitution you approve now actually runs.

    Slide text
    > windows, macos and vs code 2026-11-19

    PowerShell commands with parentheses run again on Windows

    PS> dir C:\Program Files (x86)\Git PS> [System.IO.Path]::GetFullPath('foo') blocked as command substitution since v0.63.0 ✓ run again
    packages/core/src/utils/shell-utils.ts · #30110
  17. In the macOS Seatbelt sandbox, Gemini CLI asked to trust the folder on every launch, then failed to save your answer. Now it remembers, and shell commands stop leaking terminal handles.

    Slide text
    > windows, macos and vs code 2026-11-19

    The macOS sandbox remembers which folders you trust

    trust dialog on every launch, then “Failed to save trust settings” ✓ trust is settled before Seatbelt starts ✓ each shell command stops leaking a terminal handle
    packages/cli/src/utils/sandboxFolderTrust.ts · #30164 #30184
  18. The VS Code companion bundled all of Gemini CLI's core, about 14 megabytes loaded at startup. Now it ships under 2 megabytes and starts its server only when Gemini CLI connects.

    Slide text
    > windows, macos and vs code 2026-11-19

    The VS Code companion extension is more than 80% smaller

    before 13.8 MB
    now 1.7 MB
    packages/vscode-ide-companion/src/ide-server.ts · #30181
  19. I would say yes, if you use Gemini CLI through Code Assist on Google Cloud and your sessions run long. It still isn't tuned for Gemini 4 models. The rest is in the release notes.

    Slide text
    > should you upgrade 2026-11-19

    Upgrade if your sessions run long or your network drops

    ✓ you run Gemini CLI through Code Assist on Google Cloud ✓ your sessions run long, or your network or quota gives out ✗ still not tuned for Gemini 4 models
    README.md · releases/tag/v0.69.0

What happened in AI today, in your inbox

One short email each evening. Read it and you’re caught up.

Built with ❤️ in New York