openclaw - 💡(How to fix) Fix [Feature] Configurable typing indicator TTL — 2-min default too short for long-running model calls [1 comments, 2 participants]

Problem

The Telegram typing indicator TTL is hardcoded in the bundled distribution as typingTtlMs = 2 * 6e4 (2 minutes). I see this in dist/pi-embedded-CbCYZxIb.js of the bundle shipped with [email protected].

With Claude Opus 4.x on real workloads (document analysis, multi-step planning, large-context responses), single-turn replies regularly take 3–10 minutes. Once the indicator disappears at 2 min the user assumes the bot has crashed and starts re-sending — which compounds the problem and produces stuck-session warnings server-side.

Current Workaround

We sed-patch the bundled file via cron every 15 minutes + @reboot:

sed -i 's/typingTtlMs = 2 \* 6e4/typingTtlMs = 4 * 6e4/' /usr/local/lib/node_modules/openclaw/dist/pi-embedded-*.js

This raises the TTL to 4 min. It survives across docker restart (we apply it from entrypoint.sh) but breaks on every npm install -g openclaw upgrade and has to be reapplied (we automate that, but it shouldn't be needed).

Proposed Solution

Make the typing TTL configurable in openclaw.json, e.g.

{
  "channels": {
    "telegram": {
      "typingTtlMs": 600000
    }
  }
}

Default could reasonably be raised to 5 min (300000) given current Claude model latencies — a 2-min default predates the long-context era.

Environment

• OpenClaw 2026.4.24 (commit cbcfdf6)
• Telegram channel via the bundled GrammY adapter
• Claude Opus 4.6 / 4.7 (provider anthropic-new)
• Linux container, Node.js v22.22.1

Happy to send a PR if you'd like a hand wiring this through — but I don't know the canonical path from openclaw.json schema to the bundled value, so a pointer would help. Thanks for the project!