model-thaiLLM

Auto-routing assigns Thai language tasks to the correct model by detecting keywords across code, legal, business, and academic scenarios. It matches task intent to specialized models, eliminating manual selection for better task-appropriate outputs.

OpenThaiGPT handles technical code logic, while Pathumma processes legal and government language. Typhoon summarizes business reporting, and Thalle performs academic thesis-style analysis, ensuring the right model handles each specific domain.

You need a THAILLM_TOKEN, network access to a reachable ThaiLLM API endpoint, and the configured runner script located at ~/Jit/limbs/thaillm.sh. This environment enables automatic routing and telemetry logging to token_log.jsonl.

Yes, the skill supports explicit model selection alongside automatic routing. You can bypass the keyword matching matrix and directly invoke a specific model, or list all available models to choose the best fit for your task.

Telemetry logging records API token usage to a token_log.jsonl file during execution. The runner script automatically logs consumption metrics for each invoked model, providing an audit trail for later analysis and operational tracking.

The routing matrix classifies mixed intent prompts using keyword matching across business, academic, code, and legal scenarios. It selects the dominant task category to route the request to the most appropriate specialized ThaiLLM model available.