Z.AI Gets Real-Time Tool Streaming: Watch Your Agent Think in OpenClaw 2026.2.17

If you've ever wondered what your agent is doing during those long pauses between sending a request and getting a response, OpenClaw 2026.2.17 has a treat for you. Real-time tool call streaming is now enabled by default for Z.AI providers.

What Is Tool Streaming?

Traditionally, when an AI agent decides to use a tool (like searching the web, reading a file, or executing code), you'd only see the final result. The entire tool invocation happened behind the scenes, and you'd wait in silence until everything completed.

With tool streaming (tool_stream), you can now see tool calls as they happen in real-time. This means:

• Immediate feedback — See which tool your agent is calling the moment it decides to use it
• Better debugging — Identify stuck or slow tool calls without waiting for timeouts
• Improved UX — Your users aren't staring at a blank screen wondering if something's broken
• Faster perceived response times — Even if total execution time is the same, seeing progress makes it feel faster

How It Works

Starting with OpenClaw 2026.2.17, Z.AI provider integrations automatically stream tool calls. You don't need to configure anything — it just works.

If for some reason you need to disable this behavior (perhaps for compatibility with downstream tooling that expects the old format), you can opt out per-model:

agents:
  defaults:
    model:
      primary: z.ai/your-model
      params:
        tool_stream: false

Why Z.AI First?

Z.AI's streaming architecture makes this particularly well-suited for real-time tool streaming. The protocol already supports incremental updates, so enabling tool_stream was a natural extension. Expect other providers to follow as their streaming capabilities mature.

Practical Benefits for Agent Developers

1. Multi-Tool Workflows

When your agent chains multiple tools together (search → read → summarize), you'll see each step as it happens rather than waiting for the entire chain to complete.

2. Long-Running Operations

For tools that take time (API calls, file operations, web fetches), streaming lets you show users that progress is being made.

3. Debugging Complex Agents

When an agent misbehaves, seeing the tool call stream helps you pinpoint exactly where things went wrong — was it the wrong tool choice, bad parameters, or a tool failure?

The Change in Context

This feature was contributed by @tianxiao1430-jpg in PR #18173. It's part of OpenClaw's broader effort to make agent execution more transparent and developer-friendly.

GitHub Reference: PR #18173

Upgrade: If you're running OpenClaw, update to 2026.2.17 or later to get this feature automatically. No configuration needed.

Have you noticed a difference in your agent's responsiveness since updating? Drop a comment below with your experience!