OpenClaw Digest: Realtime Voice, Tailscale Funnel Persistence, and Gateway Optimizations

The latest set of merged pull requests for OpenClaw focuses on expanding multimodal capabilities, improving infrastructure stability for self-hosted deployments, and refining the user experience across various chat integrations. Key highlights include the introduction of sophisticated realtime voice modes for Discord and critical fixes for Tailscale Funnel persistence.

Merged PRs

• fix: allow Nix store plugin hardlinks Original PR
• fix(gateway): reset webchat /new in place when dmScope is main (#77434) Original PR
• fix(failover): defer profile cooldown marking to unblock rate-limit rotation Original PR
• fix(gateway): preserve external Tailscale Funnel routes in serve mode Original PR
• feat(discord): add realtime voice modes Original PR
• fix(telegram): chain over-limit stream previews Original PR
• fix(memory): memoryFlush fires every compaction cycle instead of every other Original PR
• refactor(agents): preserve raw reasoning stream and push formatting to edge Original PR
• fix(telegram,feishu): respect reasoningDefault config for reasoning stream previews Original PR
• fix(telegram): log inbound gateway watch messages Original PR
• fix(telegram): suppress tool-only silent fallback Original PR
• Fast-path native slash commands Original PR

Key Changes

Multimodal & Channel Enhancements

OpenClaw has significantly expanded its Discord integration with the addition of realtime voice modes. This includes stt-tts, talk-buffer, and bidi modes, allowing Discord voice PCM to stream through a shared realtime provider bridge. This enables direct bidirectional consult calls and more fluid agent talkback.

For Telegram and Feishu, the system now respects the reasoningDefault configuration, ensuring that reasoning stream previews are displayed or hidden based on user preference. Additionally, Telegram now supports chaining over-limit stream previews, meaning drafts that exceed the 4,096-character limit will continue in a new message rather than simply stopping.

Infrastructure & Gateway Stability

A major pain point for self-hosted users was the overwriting of external Tailscale Funnel routes. When gateway.tailscale.mode was set to serve, restarts would clobber public Funnel configurations. The introduction of gateway.tailscale.preserveFunnel allows the gateway to detect existing Funnel routes and skip the serve command, preserving public accessibility across restarts.

In the failover system, a critical bug was fixed where rate-limit errors were blocking profile rotation due to file lock contention. By deferring the cooldown marking to a non-blocking process, OpenClaw now rotates to the next auth profile immediately, preventing multi-minute stalls in production.

Core Logic & Performance Refactors

• Reasoning Streams: Reasoning delta extraction has been refactored to preserve raw text. Formatting is now pushed to the "edge" (the channel dispatchers), which resolves a bug where thinking_delta extraction failed on same-line appends.
• Session Management: Webchat now correctly handles the /new command when dmScope is set to main, resetting the session in place rather than creating a parallel child session.
• Fast-Path Commands: Native slash commands (e.g., /status on Telegram) are now fast-pathed, bypassing the slow full-turn workspace bootstrap to provide near-instant responses.
• Memory Management: A long-standing bug in memoryFlush was resolved, ensuring it fires on every compaction cycle instead of every other cycle, improving the reliability of durable note writes.

Impact

These changes collectively reduce latency and increase the reliability of OpenClaw's agent deployments. The transition to raw reasoning streams and the implementation of fast-path commands significantly improve the perceived responsiveness of the agents.

For operators, the Tailscale and failover fixes eliminate silent failures and manual intervention requirements after restarts or rate-limit hits. For end-users, the Discord voice integration transforms the agent from a text-based bot into a real-time conversational partner, while the Telegram improvements ensure that long-form reasoning and complex drafts are delivered without truncation.