The Modelfile Manager allows you to create, edit, and build custom Ollama models using Ollama Modelfiles — without leaving VS Code.

What Is a Modelfile?

A Modelfile is a text file that defines a custom Ollama model variant. It specifies:

• Base model (FROM) — which model to build on top of
• System prompt (SYSTEM) — default behavior, persona, or instructions
• Inference parameters (PARAMETER) — temperature, context length, sampling settings
• Chat template (TEMPLATE) — custom prompt structure (advanced)
• Training messages (MESSAGE) — example conversations for few-shot prompting
• Adapters (ADAPTER) — LoRA adapter paths (advanced)

# Modelfile — code-reviewer
FROM qwen2.5-coder:7b

SYSTEM """You are an expert code reviewer. Focus on:
- Security vulnerabilities (OWASP Top 10)
- Performance bottlenecks
- Readability and maintainability
Always provide specific, actionable feedback."""

PARAMETER temperature 0.3
PARAMETER num_ctx 8192

Opening the Modelfile Manager

Click the 🦙 Ollama icon in the activity bar, then open the Modelfiles panel at the bottom of the sidebar.

Creating a New Modelfile

Click the + (New Modelfile) button in the panel header. An interactive wizard prompts you for:

1. Name — the name for the custom model (e.g., pirate-bot, strict-reviewer)
2. Base model — pick from your locally installed Ollama models
3. System prompt — describe the persona or task

The wizard creates a .modelfile file at your configured modelfiles path (default: ~/.ollama/modelfiles/<name>.modelfile), pre-populates it with the chosen settings, and opens it in the editor.

Editing a Modelfile

Click the Edit (✏) button next to any modelfile in the panel, or open the file directly. The editor provides:

• Syntax highlighting for Modelfile keywords and parameter names
• Hover documentation — hover over FROM, PARAMETER temperature, etc. to see inline docs
• Autocomplete — Ctrl+Space / Cmd+Space suggests Modelfile keywords and common parameter names

Building a Model from a Modelfile

After editing a Modelfile, build it into a runnable Ollama model:

1. Right-click the modelfile in the panel → Build Model from Modelfile
2. Or open Command Palette → "Ollama: Build Model from Modelfile"

A streaming VS Code notification shows the build progress from ollama create. The built model appears in the Local Models panel immediately after a successful build.

Modelfile Syntax Reference

FROM <model>

Specifies the base model. Required.

FROM llama3.2:3b
FROM qwen2.5-coder:7b

SYSTEM <prompt>

Sets the system message injected at the beginning of every conversation.

SYSTEM "You are a helpful assistant."

# Multi-line (triple-quoted):
SYSTEM """
You are an expert TypeScript developer.
You always write clean, type-safe code.
You prefer functional patterns over mutable state.
"""

PARAMETER <name> <value>

Overrides model inference defaults. Common parameters:

Parameter	Type	Description
temperature	float	Randomness (0 = deterministic, 1 = creative)
num_ctx	int	Context window size in tokens
top_p	float	Nucleus sampling threshold
top_k	int	Top-k sampling
repeat_penalty	float	Penalty for repeating tokens
num_predict	int	Max tokens to generate (-1 = unlimited)
stop	string	Stop sequence string

PARAMETER temperature 0.2
PARAMETER num_ctx 16384
PARAMETER stop "<|im_end|>"

MESSAGE <role> <content>

Adds example conversation turns for few-shot prompting.

MESSAGE user "What is 2+2?"
MESSAGE assistant "4."

TEMPLATE <template>

Custom Go template for prompt formatting (advanced — only needed for unusual model architectures).

ADAPTER <path>

Path to a LoRA adapter to apply on top of the base model.

LICENSE <text>

License text for the custom model.

Modelfiles Storage Location

By default, modelfiles are stored in ~/.ollama/modelfiles/. Change this with the ollama.modelfilesPath setting.

Use the Open Modelfiles Folder (📁) button in the panel header to reveal the folder in your OS file manager.

External Reference

Full Modelfile specification: github.com/ollama/ollama/blob/main/docs/modelfile.md