OpenClaw v2026.5.4-beta.1: Parallel Web Search, QQBot Fixes, and Matrix Voice Notes Land in Pre-Release

OpenClaw’s pre-release track shipped v2026.5.4-beta.1 on June 7 with a notably broad set of improvements spanning channels, model providers, and core infrastructure.

Parallel web_search integration: Parallel joins the bundled web_search provider roster alongside existing options, with PARALLEL_API_KEY discovery, guarded endpoint handling, cache-safe session IDs, and onboarding picker support. If you’ve been running web searches through a custom provider, this may be a cleaner path.

QQBot reasoning stripping: QQBot now strips model reasoning/thinking scaffolding before native delivery, preventing raw content from leaking into channel replies. Two PRs and contributions from @openperf suggest this was a real-world pain point.

MCP boundary coercing: MCP tool results now handle resource_link, resource, audio, malformed image, and future non-text/image blocks at the materialize boundary — preventing Anthropic 400 errors and poisoned session history when tools return richer content than expected.

Google Vertex recovery: After a regression that affected static catalog rows and runtime model resolution for Vertex ADC users, the latest patches restore reliability and add more robust single-provider cooldown recovery.

Matrix voice notes: Matrix now preflights voice notes before mention gating and preserves thread reads/replies through Matrix relations pagination — a meaningful quality-of-life win for Matrix-based deployments.

Auth durability: Auth profiles now live in SQLite rather than ephemeral state, official npm plugin install records keep trusted pins, and prerelease fallback integrity checks avoid carrying stale state forward.

macOS node mode: Companion app session churn reduced — node mode no longer silently reconnects away from a healthy direct Gateway session.

The full changelog runs deep on reliability and boundary-case handling, which suggests the team is hardening the platform rather than adding headline features. That’s typical of a project in this phase.