Vibeleaderboard
← All Intel
Intel / repo

gemini-cli v0.69.0 — Release v0.69.0

Source
github.com
Author
gemini-cli-robot
Date

Context

Gemini CLI 0.69.0, a release of Google's coding for the terminal, is mostly bug fixes, and the ones a developer is most likely to notice deal with the CLI waiting silently when a request to the Gemini API cannot finish. The release notes list them only as pull request titles; the details below come from the code changes and commit messages of three of those pull requests.

The first addresses network failures. Pull request 30182 targets silent loading hangs lasting several minutes when the network fails. Several behaviors added up: cancelling did not interrupt the pauses between retries, the step that picks a model retried five times before handing over to a second full retry loop, and errors that retrying cannot fix, such as an unreachable network or a certificate verification failure, were retried anyway. After the fix, cancelling stops the retries, the model-picking step gives up after two attempts, those permanent errors fail on the first try, and during network retries the loading message names the model and the attempt number even when error verbosity is set to low. The CLI also now falls back to the `ALL_PROXY` environment variable when looking for a proxy.

Two other changes cover requests that are slow rather than broken. Pull request 30187 deals with quota errors: when the API reported an exhausted quota that would not reset for hours, the CLI could keep retrying for hours. In the merged version it retries only when the wait falls within its maximum retry delay, and programs connected through the Agent Client Protocol, such as editors, now receive a message for each retry. Pull request 30183 makes the 60-second limit on waiting for the API to start responding configurable. Long-thinking requests could exceed it and be abandoned and retried. The `model.requestTimeoutMs` setting or the `GEMINI_API_HEADERS_TIMEOUT_MS` environment variable, both in milliseconds, now raise it for every sign-in method; before, the CLI took a custom value only from an experiment flag.

Developers who have seen long silent waits or slow requests restarting should find this release fails faster and says more about what it is doing. The pull requests do not give the maximum retry delay, and the release notes do not say which earlier versions had these problems.

Key takeaways · AI-distilled
  • Gemini CLI v0.69.0 lets users override the fixed 60-second API headers timeout in every auth mode, through the model.requestTimeoutMs setting or the GEMINI_API_HEADERS_TIMEOUT_MS environment variable.
  • Several fixes target long stalls: the release stops multi-hour quota retries, notifies ACP clients when a retry happens, and prevents silent multi-minute loading hangs on network failures.
  • Headless runs now clean up on abort: the CLI kills child processes, records cancelled tools and preserves the signal's exit code.
  • Custom themes can live in a standalone theme.json file, and the /model command now validates model names and accepts the model as a direct argument.
  • Memory fixes bound multi-part tool outputs, evict older inline data, and stop PDF inline data and telemetry buffers from being retained.
Terms in this piece · Glossary
  • AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
  • token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
  • multimodal — A model that works with more than text — reading images, audio, or video, and sometimes generating them too.
More from gemini-cli-robot
Recommended reads
Comments

Checking sign-in…

Loading comments…