2026-08-20 · NextClaw v0.40.0
Published: 2026-08-20 Tags: release npm minor
NextClaw v0.40.0 makes long-running work easier to resume and runtime failures easier to investigate. Large tool-heavy sessions return to the latest progress faster, frequently used Panel Apps can live in the main sidebar, QQ reconnects more reliably, and the built-in AI can follow structured log evidence across the runtime.
Enhancements
- Large sessions reopen faster. In a 44.5MB stress session with 500 tool calls in one message, time to the latest visible message dropped from about 6.94 seconds to 1.13–1.73 seconds—up to about 6× faster. The initial history response fell from 12.96MB to 261.85KB.
- Tool details load when requested. The first view shows the real tool count, tool types, and result summaries. Full arguments and results load per message when the process is expanded. A 500-call message is rendered in batches while keeping every record available.
- Panel Apps can join the main sidebar. Add an entry from the Apps list or a running App menu. Disabled Apps temporarily hide their entry, re-enabling restores it, and uninstalling or deleting an App removes it.
- Runtime failures are traceable. Structured diagnostics and
nextclaw logs querynow cover services, extensions, configuration, channels, Agents, scheduled work, tools, and network requests. The built-in AI can narrow failures by time, outcome, reason code, and correlation ID.
Fixes
- QQ reconnects after short gateway interruptions, and failed inbound submissions are no longer marked as processed, allowing a safe retry.
- Cancelled tools and Agent runs now have a distinct
cancelledterminal state instead of looking successful or becoming an undifferentiated failure.
Defaults and compatibility
- Small sessions still load complete content in one pass; tiered loading applies only to large tool histories.
- Diagnostics omit message bodies, tool arguments and results, full URLs, user identities, and credentials by default.
- Log queries report unparseable legacy lines so format drift is not mistaken for an absence of failures.
Learn more: Why large tool-heavy sessions now reopen faster