OpenClaw 3.8 Introduces Agent Traceability and Official Backup Tool Amid Rapid Release Cycle

OpenClaw has released version 3.8, a stable update that follows closely on the heels of version 3.7. The new iteration introduces significant features, including ACP traceability for AI agents and an official backup utility, openclaw backup. This release continues a rapid development pace, with version 3.7 launched on the morning of March 8th and version 3.8 on the afternoon of March 9th.

Core Enhancements in OpenClaw 3.8

The 3.8 update focuses on four primary areas: enhanced agent communication, improved data integrity, bug fixes, and security.

First, ACP (Agent Communication Protocol) traceability allows AI agents to identify the source of commands. This means agents can now distinguish "who is talking to it," enabling more granular permission control and accountability in multi-user environments. Previously, agents treated all interactions uniformly, lacking context about user identity or authorization. The new system assigns an "ID card" to each command, detailing the sender, entry point, and permissions. This feature is designed to prevent unauthorized actions and provide a clear audit trail for agent interactions.

Developer's hands typing on a keyboard, with holographic display showing 'openclaw backup' commands.

Second, the introduction of openclaw backup provides an official solution for data preservation. This tool addresses a long-standing user request for a standardized backup mechanism, particularly before major updates or configuration changes. Users can now create and verify backups using simple commands like openclaw backup create and openclaw backup verify. The utility supports options for backing up only configurations (--only-config) or excluding workspace data (--no-include-workspace). The backup files include manifest verification, ensuring integrity. The system will also prompt users to create a backup before performing potentially risky operations, such as major version upgrades.

Third, version 3.8 resolves an issue with duplicate messages in Telegram notifications. Teams using Telegram as a communication channel will no longer receive repeated alert messages, as deduplication logic has been implemented.

Finally, the update includes over 12 security fixes. While specific vulnerability details were not disclosed, the official team emphasized the importance of these patches, particularly for OpenClaw instances exposed to the public internet.

Abstract digital shield deflecting fragmented digital threats, symbolizing security patches and protection.

Additional Improvements

Beyond the core features, OpenClaw 3.8 incorporates several other enhancements:

• Configurable TalkMode Silent Timeout: Users can now adjust the duration an AI agent waits before automatically sending a response in voice mode via talk.silence Timeout Ms.

• Automatic TUI Terminal Background Detection: The command-line interface now automatically detects light terminal backgrounds and adjusts its color scheme for better visibility. Users can also force a light theme with OPENCLAW_THEME=light.

• GPT-5.4 Context Window Correction: The GPT-5.4 model in openai-codex mode now correctly utilizes a 1.05 million token window with 128k maximum output tokens, an increase from its previous context limit.

• Brave Search LLM Context Mode: Web searches can now leverage Brave's LLMContext endpoint, providing extracted summary snippets and source metadata directly, which is intended to be more compatible with AI applications.

• Podman+SELinux Compatibility Fix: For Fedora and RHEL users, the system now automatically detects SELinux mode and adds the :Z tag, streamlining permission handling.

• Slimmer Docker Image: Development dependencies and build metadata have been cleaned up, resulting in a smaller Docker image size.

Modern desk with laptop displaying code and holographic elements representing new OpenClaw features.

The rapid release schedule of OpenClaw, with frequent updates, indicates an active development community.