2026-08-10 · NextClaw v0.30.0

Published: 2026-08-10 Tags: release npm minor

NextClaw v0.30.0 is better suited to staying online on a VPS, NAS, or personal server. Unused messaging channels no longer keep separate processes resident and start only for authorization or active use. Chat also gains precise references to excerpts from earlier messages, plus better controls for images, Mermaid diagrams, and newly discovered models.

Features

• On-demand messaging channels: Disabled Weixin, Feishu, QQ, Discord, and other channels do not start separate processes. Enabling or disabling a channel while NextClaw is running automatically starts or releases its runtime without restarting the host.
• Lower idle memory: In an ARM64 Linux empty-configuration benchmark limited to 2 vCPU and 2 GiB with no active task, the three-run average working set fell from about 865–885 MiB to 164.94 MiB, a reduction of about 81%. Other architectures and active workloads will vary.
• Reference excerpts from earlier messages: Select part of a user or AI message and add it to the composer as a structured reference. The compact reference remains after sending, and the AI receives the exact snapshot captured at selection time.

Enhancements

• Images and Mermaid diagrams now share a full-screen preview with fit-to-window, zoom, pan, reset, and copy actions for images or Mermaid source.
• New-model notices stay visible at the bottom of the model picker. The larger model catalog supports search, provider filters, individual additions, and an explicit way to dismiss the current discovery batch.
• The low-frequency Service Apps shortcut has been removed from the right dock. Service Apps remain available from the Apps page and can still be pinned when needed.

Fixes

• The remove button for structured references now appears reliably on pointer hover or keyboard focus without changing the reference chip width.
• Selection controls in file previews and conversation messages appear reliably after selection and stay within the viewport.

Defaults and compatibility

• Existing messaging channels with enabled: true continue to start normally. Disabled or unconfigured channels remain stopped by default, with no configuration migration required.
• Active channels, Agents, browsers, MCP servers, and local models still add memory according to the work they perform. The ARM64 empty-configuration result is not a fixed promise for every VPS.

Learn more: Runtime resource usage · Tasks and conversation references · Model selection and configuration