OpenClaw 2026.6.5 Bundles Parallel Search, Hardens Agent Security and AI Model Recovery

OpenClawOpenClaw

OpenClaw released version 2026.6.5, integrating Parallel web search, adding a security policy for skill and plugin installations, and improving reliability for Anthropic models and MCP tool use. This update enhances the self-hosted AI assistant's ability to autonomously gather information and provides operators with more control over agent capabilities. It addresses key challenges in agentic engineering by focusing on robust tool use, security, and model integration.

OpenClaw 2026.6.5 introduces a bundled Parallel web search provider for its AI agents. It adds a security.installPolicy feature, enabling operators to approve or block skill and plugin installs. The release shifts its versioning to a YYYY.M.PATCH format.
Versioning
YYYY.M.PATCH monthly patch numbering
Web Search
Bundled Parallel provider with free/paid API options
Skill Security
security.installPolicy for operator approval
Anthropic Stability
Improved recovery for extended-thinking sessions
MCP Tool Results
Coerces non-text/image blocks to prevent errors
Matrix Channel
Voice notes preflighted, threads preserve context

These enhancements strengthen the platform's agentic capabilities, particularly for RAG (Retrieval-Augmented Generation) tasks. The new install policy provides a crucial human-in-the-loop (HITL) control point, addressing security for autonomous agents. It also improves the stability of interactions with Anthropic's extended-thinking models and the MCP (Model Context Protocol).

OpenClaw users can configure the integrated Parallel web search for their agents, with free or paid API key options. Operators can implement the security.installPolicy to manage skill installs, ensuring approved capabilities are added. The update improves recovery for Anthropic extended-thinking sessions and normalizes richer MCP tool results.

View the full update on github.com
OpenClaw🦞
OpenClaw🦞
@openclaw
X

OpenClaw 2026.6.5 🦞 🔎 Parallel web search bundled 🛡️ security.installPolicy for skill/plugin installs 🎙️ Matrix voice + thread fixes 🧠 Anthropic/MCP recovery hardened Less chaos. More claws. https://t.co/eXevfEsApf

11retweets128likes
View on X

Still wondering? A few quick answers below.

OpenClaw 2026.6.5 introduces a bundled Parallel web search provider, a new security policy for skill and plugin installations, and improved recovery for Anthropic extended-thinking sessions and MCP tool results. It also includes fixes for Matrix voice and thread functionalities.

The security.installPolicy allows OpenClaw operators to approve or block skill and plugin installations using a trusted local command. This feature runs after staging and is designed to fail closed if enabled but unavailable, enhancing control over agent capabilities.

OpenClaw 2026.6.5 hardens recovery for Anthropic extended-thinking sessions, ensuring better stability after prompt-cache expiry or Gateway restarts. It also normalizes richer MCP tool blocks, preventing malformed content from affecting session history.

OpenClaw now uses a YYYY.M.PATCH release naming convention. The patch number is a monthly counter, not a calendar day, designed to support future critical fixes on prior releases while maintaining compatibility with old tags.

Parallel is now a bundled web search provider within OpenClaw. It offers a zero-config free path or supports a paid API key for dense excerpts optimized for LLM context, allowing agents to perform web searches.

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

Share this update