Skip to content

Version 1.6.0

Choose a tag to compare

@Unknnownnn Unknnownnn released this 18 Jun 11:39

Destructive Action Security & Consent Overhaul

Prior to v1.6.0, the AI agent could not perform destructive operations (such as deleting files or altering system states) due to missing execution paths and verification logic. Whenever the AI initiates a destructive or sensitive command, the system prompts for consent.

Local File Deletion (delete_local_file)

  • Added a secure file-purging tool to the agent's action schema inside SanitizationSandbox.
  • Includes path traversal jailguards to ensure deletions are restricted to safe, authorized user folders
  • Prompts the user for absolute permission prior to executing os.remove.

Automatic Path & Username Resolution

  • Adjusted application launchers and file tools to automatically resolve user directory placeholders. Commands targeting paths like Downloads/Other/ or referencing foreign system usernames are dynamically rewritten to match the active PC host's directory format.

Core & Session Updates (Since v1.5.2)

  • AI Chat History Persistence: Locally persists the last 15 chats across sessions, preventing context loss when relaunching the client.
  • NLP Command Retry: Added a retry option for the last NLP command if execution encounters errors.
  • STUN and UDP Hole Punching Fallbacks: Fixed connection deadlocks by introducing WebRTC-inspired fallback routing for remote control sessions.