openclaw - 💡(How to fix) Fix [Bug]: can't get "start talk" to work with a local whisper server . [1 comments, 2 participants]

Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

AI generated , The "openai" realtime voice provider is reported as not configured even when its STT baseUrl and apiKey are explicitly defined within the plugins configuration block.

Steps to reproduce

AI generated , 1. Configure plugins.entries.openai.config.stt in openclaw.json with a valid baseUrl (e.g., a local Whisper server IP) and apiKey. 2. Restart the OpenClaw Gateway to apply the configuration. 3. Attempt to start a Talk mode session. 4. Observe the error: Realtime voice provider "openai" is not configured.

Expected behavior

The Gateway should successfully initialize the "openai" voice provider using the baseUrl and apiKey specified in the plugins.entries.openai.config.stt block, allowing Talk mode to route STT requests to the configured endpoint instead of reporting it as unconfigured.

Actual behavior

The Gateway fails to initialize the voice provider despite a schema-valid configuration in openclaw.json, resulting in the user-visible error: Error: Realtime voice provider "openai" is not configured during Talk mode session attempts.

OpenClaw version

2026.4.26

Operating system

ubuntu 24.04 ?

Install method

your openclaw install

Model

all local via ollama , gemma , qwen3.5 and qwen3.6 models .

Provider / routing chain

openclaw on VM to gateway .

Additional provider/model setup details

• Infrastructure: Local Whisper server running on AMD hardware via insanely-fast-whisper-rocm.
• Server Endpoint: http://192.168.0.145:8000/v1 (verified reachable via /health endpoint).
• Config Context:
  • plugins.entries.openai.config.stt was configured with the local baseUrl and a placeholder apiKey.
  • A local skill folder /home/peter/.openclaw/workspace/skills/local-whisper was created to avoid config clobbering.
  • Attempted overrides in tools.voice and tools.media.audio were either ignored by the system or stripped by the Gateway during restart due to schema validation.
• Observed behavior: The system consistently fails at the provider resolution stage, ignoring the provided baseUrl and defaulting to a "not configured" state for the "openai" provider.

Logs, screenshots, and evidence

**Logs & Evidence:**

**1. User-Visible Error:**
`log ⇄ res ✗ talk.realtime.session 19ms errorCode=UNAVAILABLE errorMessage=Error: Realtime voice provider "openai" is not configured conn=81c7a63d…4a15 id=e5e849bc…a15e`

**2. Verified Server Reachability:**
`root@ave:~# curl http://192.168.0.145:8000/health`
`OK`

**3. Configuration Applied (in `openclaw.json`):**

"plugins": {
  "entries": {
    "openai": {
      "enabled": true,
      "config": {
        "stt": {
          "enabled": true,
          "provider": "openai",
          "baseUrl": "http://192.168.0.145:8000/v1",
          "apiKey": "***"
        }
      }
    }
  }
}


**4. Validation Evidence:**
The configuration was successfully accepted by the Gateway during reload (not clobbered/restored from backup), confirming that the `plugins.entries.openai.config.stt` structure is schema-valid, yet the provider remains "not configured" at runtime.

Impact and severity

• Affected users/systems/channels: Users attempting to use Talk mode with a local or custom STT (Speech-to-Text) provider via the openai plugin configuration.
• Severity: Blocks workflow.
• Frequency: Always (constant failure during provider resolution).
• Consequence: Total loss of Talk mode functionality; users are unable to use voice-to-text capabilities even when a functional local server is available and correctly referenced in the configuration.

Additional information

NOT_ENOUGH_INFO