A diagnostic framework for validating and debugging local OpenAI-compatible speech-to-text servers within an AI agent environment.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install local-stt-workflow
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install local-stt-workflow using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Local STT Workflow?
The Local STT Workflow is a specialized diagnostic skill designed to troubleshoot and optimize local speech-to-text servers, specifically those providing OpenAI-compatible endpoints like MLX-audio. By following this workflow, developers can systematically isolate issues between the inference server and the application routing layer. Utilizing Openclaw Skills for audio debugging ensures that multipart uploads, model path fallbacks, and streaming behaviors are correctly configured for high-performance local transcription.
This skill is particularly valuable for developers working with local model deployments where standard cloud-based debugging tools are unavailable. It provides a blueprint for verifying the full transcription path, from health checks to granular timestamp validation, ensuring that Openclaw Skills integrate seamlessly with local hardware acceleration.
Local STT Workflow Use Cases
- Troubleshooting cases where an AI agent fails to generate transcripts from audio attachments.
- Validating the registration and availability of local models like Qwen3-ASR.
- Testing multipart form upload compatibility for various audio containers such as WAV, MP3, or M4A.
- Debugging custom server parameters like chunk duration and frame thresholds that differ from standard OpenAI specs.
How Local STT Workflow Works
- Perform a basic server health check and model inventory to ensure the STT service is responsive.
- Isolate the server by sending a direct curl request to the transcription endpoint to prove the inference engine works independently of the client.
- Distinguish between server-side failures (inference bugs) and client-side routing failures (OpenClaw configuration issues).
- Inspect the request shape by comparing client-sent multipart fields against the local OpenAPI document.
- Verify advanced output formats and streaming SSE events against the local API reference documentation.
- Conduct an integration test within the AI agent to ensure the media tools are correctly dispatching audio files to the local base URL.
Local STT Workflow Setup
Ensure your local STT server is active, typically at http://127.0.0.1:8000. You can verify the setup using these bash commands:
# Check server health and registered models
curl http://127.0.0.1:8000/health
curl http://127.0.0.1:8000/v1/models
# Perform a manual transcription test
curl -X POST http://127.0.0.1:8000/v1/audio/transcriptions \
-F [email protected] \
-F model=qwen3-asr \
-F response_format=json
If the STT task is not active, start the server with the appropriate environment variables:
MLX_AUDIO_SERVER_TASKS=stt uv run python server.py
Local STT Workflow Data Schema & Taxonomy
The workflow interacts with a multipart/form-data schema. Below are the primary fields managed by this skill when interfacing with Openclaw Skills:
| Field | Type | Required | Description |
|---|---|---|---|
| file | file | Yes | The raw audio binary data. |
| model | string | Yes | The ID of the local model (e.g., qwen3-asr or a local path). |
| response_format | string | No | Format of the transcript (json, text, verbose_json, srt, vtt). |
| stream | boolean | No | Enables SSE events for real-time transcription. |
| chunk_duration | float | No | Specific to local MLX servers for prefill control. |
Local STT Workflow Advanced Features
- Support for local model-path fallback when a model is not found in the global registry.
- Detailed word-level and segment-level timestamp granularity for precise alignment.
- Server-Sent Events (SSE) validation for low-latency streaming transcription.
- Diagnostic rules for identifying niche input container bugs (e.g., M4A vs WAV compatibility).
- Direct OpenAPI schema comparison to identify non-standard field support like prefill_step_size.
SKILL.md
Loading
Related Openclaw Skills
China Briefing
A sophisticated AI skill for generating multi-dimensional Chinese news reports ranging from global trends to hyper-local street updates.
China Resources Information Assistant (CRC)
A dedicated information assistant for China Resources to search products, news, financials, and official corporate resources.
Skill HR
A comprehensive people operations manager for your AI agent ecosystem that handles job intake, skill matching, and performance logging.
B2B SDR Agent
An open-source AI Sales Development Representative template designed to automate the entire B2B sales funnel across multiple communication channels.
PR Specialist
An AI-driven public relations consultant that provides expert guidance on media relations, press coverage, and reputation building.
Lark Enterprise Agent Skill
A comprehensive automation suite for the Lark (Feishu) ecosystem that integrates AI agents with documents, multidimensional tables, and knowledge bases.
Featured*
