Skip to content
Agent Radar

Laya guard 0.4.0 stops outages from disabling the guard

Updated Laya guard

Summary

Laya guard v0.4.0, released October 13, 2026, now asks or blocks unjudged tool calls when its server is down and adds a login service and alerts.

Tool
Laya guard
Category
Coding agents
Date
Open source
Yes
License
MIT
Pricing
Not stated in the sources we cited.
Sources

Laya guard v0.4.0 was published on October 13, 2026 by scs0209 on GitHub. The release notes say an outage no longer turns the guard off: when the server is down, calls it cannot judge are asked about or blocked instead of passing.

Release facts

FactDetail
ProductLaya guard, a local guardrail for AI coding agents (per the repository description)
Versionv0.4.0
Release dateOctober 13, 2026
LicenseMIT, as reported by the GitHub API
Where to get itpip install -U laya-guardrail (the PyPI name; the command is laya-guard)

What changed

  • Safer default on outage. LAYA_GUARD_FAIL now defaults to ask instead of open. A call the offline check cannot judge is no longer passed. Claude Code, Cursor, Qwen Code, Copilot CLI and Amp's permissions helper ask you; Gemini CLI, Codex, Windsurf, Cline, OpenCode, Kilo and the Amp plugin have no ask lane, so they block it. LAYA_GUARD_FAIL=open restores the old behaviour.
  • Start-at-login service. laya-guard service install | uninstall | status sets up a macOS LaunchAgent or a Linux systemd user unit running laya-guard serve, restarted after a crash. Windows is not supported yet.
  • Degraded indicator. The dashboard turns amber ("Degraded") when calls in the last hour were judged by the offline check or after a model error, and each call shows who judged it.
  • Alert hook. LAYA_GUARD_ALERT_CMD runs a shell command in the background when calls are judged without the model, once per outage kind per LAYA_GUARD_ALERT_COOLDOWN seconds (default 600).
  • Fixes. The Gemini CLI, Qwen Code, Cline and Kilo Code adapters now give the offline check the bare shell command, so git status is no longer counted as unjudged with the server down. laya-guard serve also exits with an error right away if something already answers on the port.

Questions

Does the guard still pass calls when the server is down? Not by default. According to the release notes, calls the offline check cannot judge are asked about or blocked, unless you set LAYA_GUARD_FAIL=open.

How much memory does the service use? The release notes say --no-warmup loads the model on the first call, using about 90 MB idle instead of 1.4 GB.

Does it run on Windows? The release notes say Windows is not supported yet for the service.

Browse the Laya guard hub, the coding agents category, or the related Cline SDK v0.0.93 launch.

Questions

What is Laya guard?

According to its repository description, it is a free local guardrail for AI coding agents that judges every tool_call on-device (regex + Laya) as allow, review or block.

Is Laya guard open source?

The GitHub API reports an MIT license for the repository.

Filed on Agent Radar. Claims in this note are limited to the sources above.