Provider Modes provider_modes

Use to specify how the prompt would be constructed for the LLM, you can modify or add your own as needed in the config.yaml - provider_modes.

mode controls how custom instructions/input (system/user) are arranged for the provider, ex:
- plain → instructions and input without extras
- flat → simple “System/User/Assistant” text blocks with headers
- chatml → [{role, content}, …]
- mistral-instruct, llama2-chat, anthropic → provider-specific wrappers

For Ollama use raw: true to bypass standard provider prompt formatting and use your custom prompt mode.

More detailed examples

There are two fields available that are used in the template prompts. It could be instructions for system instructions and input as a user query.

plain

{% if instructions %}{{ instructions }}\n{% endif %}{{ input }}

Example

You are an assistant.
Write a small tech article.

flat

{% if instructions %}### System:
{{ instructions }}

{% endif %}### User:
{{ input }}

### Assistant:

### System:
You are an assistant.

### User:
Write a small tech article.

### Assistant:

chatml

[
  {% if instructions %}{"role": "system", "content": "{{ instructions }}" },
  {% endif %}{"role": "user", "content": "{{ input }}" }
]

[
  {"role": "system", "content": "You are an assistant."},
  {"role": "user", "content": "Write a small tech article."}
]

anthropic

Human: {% if instructions %}{instructions}

{% endif %}{{ input }}

Assistant:

Human: You are an assistant.

Write a small tech article.

Assistant: