OpenClaw

GPT-5.6 support landed alongside openclaw attach, a new command for resuming external harness sessions against active Gateway connections.

Telegram gained Codex integration. Telegram can now start Codex pairing with /login, steer active Codex runs, and recover final replies across transient API failures. Rich HTML delivery — tables, lists, expandable blockquotes, and sticker paths — was also strengthened.

Event-driven scheduling arrived. The on-exit schedule kind wakes an agent when a watched command exits; session-targeted runs can detach cleanly.

Channel control expanded. Slack relay mode, native Mattermost /oc_queue, and per-DM model overrides give operators more fine-grained automation. openclaw agent --message-file and the RAFT CLI wake bridge add file-driven and remote wake-up paths.

New models and faster talk mode landed. GPT-5.6 joined the catalog. Fast mode for short conversational turns enables lightweight responses, reverting automatically for longer runs. Zai model synthesis, GLM overload failover, and native reasoning-level selection follow the active catalog more consistently.

Native apps refreshed. iOS adopted the iOS 26 visual system with clearer Chat, Talk, onboarding, and navigation flows. Android settings detail panels improved configuration visibility.

Agent reliability and security hardened. Codex partial deltas, harness activation, and long-context prompt-cache stability reduce lost progress. Security boundaries tightened across sandbox binds, MCP stdio, Codex HTTP access, and elevated sender checks.