OpenClaw Update 5.26 is quite UNDERRATED (Speed Upgrade)

OpenClaw 5.26 cuts startup scans and splits reply delivery, making the agent feel materially faster with zero config edits.

## Startup rewrite mirrors last week's Hermes Agent speed push The 5.26 release borrows the same playbook the creator saw in the recent Hermes Agent upgrade: multiple PRs targeting boot performance. Previously, every OpenClaw launch rescanned plugins, channels, active sessions, usage costs, warning states, scheduled services, and file system states. Now the agent caches what it already knows and only refreshes what changed. On a busy install or a VPS with a slower disk, that redundant work was the bottleneck.

## Replies no longer wait on background tasks The second big win: the user-facing send is now decoupled from slower jobs like follow-up searches, memory writes, and cost-log flushes. The message goes out immediately; the housekeeping runs after. End result is less time staring at an indefinite typing indicator on Discord or Telegram. Active sessions also see less memory thrashing on hot code paths, so the agent stays responsive longer between restarts.

## Transcripts become the unified backbone Transcripts were a side feature; in 5.26 they consolidate meeting summaries, provider chunks, media provenance, cleaned-up user turns, web chat replies, and even CLI replays into a single pipeline. Sessions replay more consistently across the web UI and the terminal.

## Channel hardening across Telegram, WhatsApp, iMessage, Discord Telegram gets better typing-indicator and forum-topic handling, iMessage gets stronger attachments and duplicate-message fixes, WhatsApp group and media behavior is restored, and Discord voice playback and model picking are smoother. Reactions are finally wired correctly across Signal, iMessage, and WhatsApp. Real-time talk sessions are now inspectable from the web UI and Discord voice, letting you steer mid-conversation or cancel a runaway response. Wake-name detection triggers less often when idle.

## Tightened runtime guardrails Browser tooling gets better SSRF protection so internal endpoints stay fenced off. Tool-call scrubbing is safer for log/transcript exports, stale device tokens are rejected immediately, and stream-response content boundaries are cleaner. The attack surface shrinks with no user-visible behavior change.

## Quality-of-life additions Named auth profiles let you switch model credentials without editing raw config. A new activity tab in the control UI shows agent state at a glance. Codex recovery is more resilient on edge cases, and Android/iOS onboarding is slightly improved.