NAME

MODEL ALIAS — inline model switching and short-name aliases

SYNOPSIS

@NAME [prompt...]

/model alias [list|add ALIAS FULLNAME|set ALIAS FULLNAME|remove ALIAS]

DESCRIPTION

Harvey supports two ways to work with multiple models in a session:

@NAME syntax — switch the active model inline, as part of a prompt. /model alias — define short names for long model identifiers.

Both are preserved in the session recording as Fountain notes so the memory miner can attribute turns to the correct model.

AT-MENTION SWITCHING

Prefix any prompt with @NAME to switch to that model for this turn and all subsequent turns:

@phi-mini summarise this function in one sentence

@qwen-coding rewrite the loop to avoid the allocation

If NAME matches a registered llamafile model, Harvey stops the current server and starts the new one. If NAME matches an Ollama model, Harvey switches the Ollama client. If NAME is not recognised, the @ prefix is treated as part of the normal prompt — no error, no switch.

Conversation history is preserved unchanged across a switch. The model switch is recorded in the session file as:

[[model switch: NAME (BACKEND) at TIMESTAMP]]

The session title page also records the starting backend:

Backend: llamafile

Use @NAME with no trailing text to switch model without sending a prompt:

@phi-mini

MODEL ALIASES

Aliases let you use short, memorable names for long model identifiers. They are stored in agents/harvey.yaml under model_aliases: and persist across sessions. Aliases resolve at the @NAME lookup step, so:

/model alias add coder qwen2.5-coder:7b

@coder tell me about this function

is equivalent to switching to qwen2.5-coder:7b.

SUBCOMMANDS

/model alias list List all defined aliases and their full model names.

/model alias add ALIAS FULLNAME Define a new alias. FULLNAME is an Ollama model identifier or a registered llamafile name. add is the preferred form; set is also accepted for compatibility.

/model alias set ALIAS FULLNAME Alias for add.