LM Studio has announced the introduction of the "Auto Review" feature to Bionic, which automatically reviews commands that agents attempt to execute. The company stated that this will reduce the burden on users to manually review and approve every command.
Bionic Launches Auto Review Feature to Automate Command Approval
This article is a translation. Read the Japanese original
Under the new system, commands are first sent to a deterministic parser called the "Shell Judge." This parser does not use an LLM; instead, it parses the command into an Abstract Syntax Tree (AST) to extract its functionality and matches it against a set of known safe commands.
Commands deemed safe by the Shell Judge are granted immediate permission for execution. For other commands, a second-stage review sub-agent called the "Shell Reviewer" verifies the session history to determine whether execution should be permitted.
Reviewing every command via an LLM presented the challenge of increased costs. According to the company, there are cases where mechanical analysis by the Shell Judge alone enabled the automatic approval of up to 82% of the commands executed by agents.
Source: How Auto Review works in Bionic (LM Studio Blog, 2026-08-27)