OpenClaw v2026.7.2-beta.3 Lands with Session Rewind, MCP Apps, and State Safety Overhaul

OpenClaw v2026.7.2-beta.3 landed on July 18, 2026, the latest in a rapid series of releases hardening the platform for production use. The release spans five major areas of work: state safety, durable channel delivery, session rewind, interactive MCP apps, and expanded questions and approvals.

State Safety and Recovery

The most critical infrastructure work in this release: protecting persisted data against primary-database damage. The new quarantine store survives database corruption events. Paired with this are crash-recoverable SQLite snapshots, crash-durable filesystem publication, schema-upgrade data-loss rejection, and rollback-writer snapshot recovery. Contributors @vincentkoc led the effort across five PRs.

Durable Channel Delivery

Messages accepted across 10+ platforms — Telegram, Signal, Slack, QQBot, Twitch, Synology Chat, Tlon, IRC, and Zalo User — are now kept recoverable across gateway restarts and local crashes via a shared ingress drain and dead-letter recovery. This addresses a longstanding reliability gap for users running OpenClaw as a always-on personal assistant.

Session Rewind and Branching

Conversations can now be rewound or forked from any individual message. Users can switch transcript branches across web and native apps, fork upstream Codex sessions, and restore prompt images after a fork. Branch-safe queued sends are preserved, and stale-pane writes are rejected. This enables more experimental multi-agent workflows without losing conversation history.

Interactive MCP Apps and Dashboards

MCP Apps — ticketed MCP apps with bound tools, resources, and bounded context updates — can now be opened from channel replies and pinned to durable dashboards. The shared sandbox has been hardened, and native plugins can declare MCP Apps directly. This brings OpenClaw closer to a true multi-app agent platform rather than a single-task automation tool.

Questions and Approvals Everywhere

The questions and approvals system has been significantly expanded. Agents can now ask structured questions with option cards across web, channels, macOS, and native apps. Approvals gain push notifications, history, fair queuing, headless resolution, Claude tool-request relay, reviewer detail, and clearer formatted prompts — a major quality-of-life improvement for users delegating sensitive actions to their agents.

Full release notes are available on GitHub.