Tool calling with local models

• Distinguish reasoning failure, parser failure, and permissions failure.
• Test tools with a minimal case before using real repos.
• Choose when to use Ollama, Open WebUI, Hermes, MCP, or vLLM.

Layered diagnosis

• Model: does it know how to follow JSON/tool format?
• Runtime: do Ollama, Hermes, Open WebUI, or vLLM expect the same format?
• Parser: does it extract arguments or stick with free text?
• Permissions: does the tool exist and can it run?

Prueba mínima:
Herramienta disponible: get_time({ "timezone": "Europe/Madrid" })

Pregunta:
"Usa la herramienta get_time para decirme la hora en Madrid.
No inventes la hora. Si no puedes llamar la herramienta, dilo."

Resultado esperado:
- llamada estructurada a get_time
- argumentos válidos
- respuesta basada en salida real

Common failures

• The model describes the tool in text but doesn't invoke it.
• It generates invalid JSON or with made-up fields.
• The runtime hides the error and it seems like "nothing happens".
• A quantization or conversion has degraded tool behavior.
• The tool executes arbitrary code without human review.

Checklist:
- Log de prompt completo
- Log de tool schema
- Log de salida del modelo antes del parser
- Log de argumentos parseados
- Log de ejecución real
- Regla de aprobación humana para escritura

Official sources

• Ollama API
• Open WebUI: Tools & Functions
• MCP specification: tools
• Hermes Agent documentation