OpenClaw 2026.7.2-beta.7: durable state and delivery, session rewind/fork, Claude Opus 5 and Kimi K3
Large beta covering crash recovery for persisted state, durable message delivery across nine chat platforms, and conversation rewind/fork across web and native apps. Adds Claude Opus 5, Kimi K3, and GPT Live realtime support — but realtime Talk now requires a supported OpenAI Platform API key, as the Codex OAuth fallback is rejected.
PUBLISHED2026-08-02
OBSERVED2026-08-11
AGE9d
SOURCES1
- State safety: quarantine store surviving primary-database damage, crash-recoverable SQLite snapshots, crash-durable filesystem publication, schema-upgrade data-loss rejection, rollback-writer snapshot recovery
- Durable channel delivery via shared ingress drain and dead-letter recovery for Telegram, Signal, Slack, QQBot, Twitch, Synology Chat, Tlon, IRC, Zalo User — accepted messages survive gateway restarts and local crashes
- Session rewind and branching: rewind or fork from individual messages, switch transcript branches on web and native, fork upstream Codex sessions, branch-safe queued sends, stale-pane write rejection, prompt images restored after a fork
- Models: Claude Opus 5 added across catalog and runtime, Kimi K3, GPT Live realtime
- Realtime Talk now requires a supported OpenAI Platform API key — the Codex OAuth fallback is rejected. OpenAI and Gemini video added to Talk
- MCP Apps: ticketed host with bound tools/resources and bounded context updates, opened from channel replies, pinned to durable dashboards, hardened shared sandbox, manifest-declared apps for native plugins
- 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
- Meetings: Teams, Zoom and Google Meet guest plugins enabled by default with durable transcript collection
- Wear OS companion (phone-proxied) with agent/session/model selection, Talk controls, audio-reactive playback, instant-talk tile
- Guided setup and local inference: local-provider detection during onboarding, in-process RAM-gated llama.cpp/Gemma GGUF path, Baseten Model API, live provider catalog discovery, model downloads from web and macOS setup, lean mode, memory imports
- Memory: fast active-memory recall, cross-conversation recall on by default for personal installs, guided imports from Claude Code/Codex/Hermes, dedicated settings page
- Scheduling: per-job dynamic cadence, gated script payloads, durable schedule-source streaming, cron-backed heartbeat monitors, heartbeat-task conversion,
/loop - Channels: Buzz plugin, Slack user-identity and Agent View modes, Telegram rich blocks and native Markdown lists, richer Matrix formatting; per-tab browser copilot and batch browser CLI
COMMUNITY
No curated reactions recorded for this event. Facts and takes are kept in separate layers — community context is added by hand, never blended into the record above.