Problem

With the default streaming: "partial" on Telegram, tool-heavy agents (multi-agent setups with frequent exec, config, sessions_send calls) create a poor UX:

1. Agent starts generating text → preview message appears in Telegram
2. Agent calls a tool → preview freezes (looks like the agent hung)
3. Tool runs for 5-60 seconds → zero feedback to the user — no typing indicator, no update
4. Agent continues → text either updates the old message (user already read and scrolled past) or appears as a disconnected fragment

This is especially bad for orchestrator agents that make 3-10 tool calls per turn.

Solution that works

Switching to block streaming + instant typing eliminates the problem:

{
  "agents": {
    "defaults": {
      "blockStreamingDefault": "on",
      "blockStreamingBreak": "text_end",
      "blockStreamingCoalesce": {
        "minChars": 200,
        "maxChars": 3000,
        "idleMs": 3000
      },
      "typingMode": "instant"
    }
  },
  "channels": {
    "telegram": {
      "streaming": "off"
    }
  }
}

Now the user sees: typing... → complete text block → typing... (during tool call) → next text block. No more frozen previews.

Suggestion

• Document this pattern in the multi-agent or streaming docs as a recommended config for tool-heavy setups
• Consider making blockStreaming + instant typing the default when an agent has many tool calls, or at least mention the tradeoff in the wizard/doctor output
• The partial streaming mode works great for simple chatbots but misleads users in complex agent scenarios

Environment

• OpenClaw 2026.3.13
• Telegram channel
• 14 agents, tool-heavy (exec, config, sessions_send, cron, etc.)