OpenClaw v2026.3.7: Authentication Fixes, Brave LLM Search, & macOS Polish

The OpenClaw v2026.3.7 update has officially landed. While largely a maintenance and refinement release following the massive features of earlier updates (like the Native PDF Intelligence in v2026.3.2), this version resolves several critical deployment friction points and adds exciting search capabilities.

Critical Authentication Configuration Fix

For self-hosters or those migrating configurations, v2026.3.7 directly addresses a frustrating initialization error.

Previously, if both auth.token and auth.password were present in the configuration without explicitly stipulating an auth.mode, the gateway could fail to start. This release forces users to set gateway.auth.mode to either "token" or "password" in their openclaw.json file. This change ensures safer predictability for setups migrating existing authentication methods.

For more details on securely configuring your gateway, review our OpenClaw Security Best Practices.

Brave Web Search: LLM Context Integration

A major addition for power users is the new opt-in mode for the Brave web search tool: tools.web.search.brave.mode: "llm-context".

This configuration allows the web_search component to tap directly into Brave's LLM Context endpoint. Rather than just returning raw HTML or scraped text, it provides extracted grounding snippets enriched with source metadata. This dramatically improves the contextual accuracy of agents conducting independent research.

(Note: Ensure your Search plan is active, as recent documentation updates clarify the transition away from defunct legacy plans).

Enhanced Context Windows: GPT-5.4 Compatibility

OpenClaw 2026.3.7 boasts serious forward-compatibility for massive context tasks. Support for GPT-5.4 now fully utilizes the staggering 1,050,000-token context window (with 128,000 max output tokens), breaking free from older Codex limitations. If you thought the 1 Million Token Context Window was impressive, the improved stability here makes document analysis even smoother.

(Note: There are current community discussions around google-vertex/gemini-3.1-flash-lite-preview compatibilities, which may be addressed in future minor patches).

macOS Fixes and TUI Polish

macOS users receive a substantial amount of polish in this release:

Service Recovery: openclaw update can now recover elegantly from a disabled gateway service by re-enabling LaunchAgent services before the updater bootstrap.
Chat UI Quality of Life: The macOS app now preserves plain-text paste semantics and strips out noisy wrapper debug traces from transcripts.
Universal Binaries: Mac release packaging now defaults to universal binaries (arm64/x86_64) ensuring peak native performance across Apple Silicon and Intel machines.

Additionally, the Terminal User Interface (TUI) is smarter. When launched inside a configured agent workspace, it will automatically infer the active agent from that context.

Need help setting up these new configurations? Browse the Complete OpenClaw Deployment Guide.