tmf save — Save Template

Syntax

tmf save <name> --baseUrl <url> --key <api-key> [options]

Description

tmf save creates a reusable parameter template that can later be invoked by tmf use <name>. Templates are ideal for:

• Frequently-used provider configurations (e.g. your self-hosted Ollama setup)
• Sharing configurations across team members (templates are stored in ~/.tmf/templates/)
• Quick switching between different models or endpoints without re-entering parameters

When a template exists with the same name as a provider, tmf use will apply the template instead of querying the Provider API.

Options

Option	Description
--app <app>	Target application (optional — can be overridden at use time)
--baseUrl <url>	Required. Base URL for the API endpoint
-k, --key <api-key>	Required. API key
-m, --model <model>	Single model name
--models <models...>	Multiple models: opus=X,sonnet=Y,haiku=Z or positional
--env <env...>	Custom env vars to set (KEY=VALUE format)
--effort <level>	Effort level: low, medium, high, xhigh

Examples

# Save a template for a self-hosted Ollama instance
tmf save ollama --baseUrl http://localhost:11434/v1 --key ollama --model llama3

# Save a DeepSeek template with role-based models for Claude Code
tmf save deepseek --app claude-code --baseUrl https://api.deepseek.com/v1 --key sk-xxx \
  --models opus=deepseek-v4-pro,sonnet=deepseek-v4-pro,haiku=deepseek-v4-flash

# Save a generic OpenAI-compatible template
tmf save myproxy --baseUrl https://proxy.example.com/v1 --key sk-xxx

Using Templates

Once saved, templates are automatically matched by tmf use:

# Apply the "ollama" template to all detected apps
tmf use ollama

# Apply the "deepseek" template (overrides --app if template has one)
tmf use deepseek

# Apply template but override the model at runtime
tmf use deepseek --model deepseek-v4-flash

CLI options (--key, --model, --models, --env, --effort, --app) take precedence over template values.

Template Storage

Templates are stored in ~/.tmf/templates/ as a JSON file. You can edit them directly or share them across machines.