OpenClaw v2026.5.26: Faster Startup, Core Transcripts, and Production-Ready Channels
The OpenClaw team shipped a major release today (May 27, 2026), packed with reliability improvements across the board.
Faster Gateway, faster replies. The release eliminates repeated scans at startup — plugin, channel, session, usage-cost, warning, scheduled-service, and filesystem checks are now cached. Visible replies also separate user-facing sends from slower follow-up work, making the experience feel snappier.
Transcripts are now core. Meeting summaries, source-provider chunks, cleaned user turns, media provenance, Codex mirrors, WebChat replies, and CLI/TUI replay now all route through a single, more reliable transcript path. This is a meaningful architectural shift.
Channels hit production-ready status. Telegram gains typing/progress context and forum topics. iMessage handles attachment roots, remote media staging, and duplicate local sources. WhatsApp restores group/media behavior. Discord improves voice playback and model picking. Signal and iMessage also get reaction approvals.
Better voice and Talk. Realtime Talk runs can now be inspected, steered, cancelled, or followed up directly from the Web UI and Discord voice. Wake-word handling is more tolerant of ambient speech without loosening security.
Security hardening throughout. Browser snapshot reads honor SSRF policy, system-event text cannot spoof nested prompt markers, fetched file text is wrapped as external content, ClickClack inbound sender allowlists run before agent dispatch, stale device tokens are rejected, and serialized tool-call text is scrubbed from replies.
Providers and local models are steadier. Notable improvements: named auth profiles, OpenAI sampling params, Codex app-server resume/timeout/usage-limit recovery, dynamic tool-schema guards, xAI usage-limit surfacing, Ollama top-p normalization, and local approval resolution.
Install and update paths hardened. Alpine installs, trusted runtime fallback roots, stable update channels, Docker/package timeouts, Windows Scheduled Tasks, and macOS runner bootstraps all received attention.
The release also ships better observability: an Activity tab, gateway secret-prep traces, tool/model stream progress, explicit fast-mode status, systemd Gateway hygiene, OpenTelemetry LLM spans, and richer telemetry signals.
What’s the story of the day? The transcript pivot feels like the most significant architectural shift — it means OpenClaw is building around conversation context as a first-class concept, not an afterthought. Combined with the channel production-readiness, this release signals the project is maturing from “works well for enthusiasts” to “reliable enough for real teams.”