Model Selection & Token Modes

Start With The Provider Default

For most users, the best first step is simple:

1. choose a provider
2. keep its default model
3. work with it for a while
4. only then optimize for price, speed, or reasoning depth

Current provider defaults are:

• OpenRouter: openai/gpt-oss-120b:free
• OpenAI: gpt-4o
• Anthropic: claude-sonnet-4-20250514

Change The Model In The TUI

Use:

/model

This opens the interactive selector for provider and model choice during a live session.

Set A Default Model In Config

ai:
  provider: openai
  model: gpt-4o

If you omit model, YeePilot uses the provider default.

Override A Model For One Run

yeepilot exec --provider openai --model gpt-4o "summarize changed files"

This is ideal when you want a stronger or cheaper model for one task without changing your day-to-day default.

Token Modes

YeePilot supports two main token-use styles:

• normal
• saver

Switch during a session:

/tokenmode normal
/tokenmode saver

Set it in config:

ai:
  token_mode: saver

What normal is for

• richer context retention
• fuller answers
• the best default for everyday use

What saver is for

• long sessions
• cost-sensitive work
• smaller, tighter answers
• reducing context pressure

Usage And Efficiency Controls

During a session, these commands help you understand and tune model usage:

/tokens
/efficiency

/tokens shows current usage context. /efficiency exposes the adaptive saver behavior that can reduce follow-up calls and verbose output when a session becomes expensive or token-heavy.

The related config is:

ai:
  efficiency:
    mode: adaptive
    policy: soft

For most users, adaptive with soft policy is the safest balance: YeePilot can reduce waste without suddenly changing the whole workflow.

Other Model-Related Settings

These settings also influence how a model behaves in real usage:

ai:
  max_tokens: 4096
  conversation_max_history: 10
  output_truncate_length: 500
  max_response_length: 102400
  usage_warning_tokens: 0
  usage_warning_cost: 0

Use lower values when cost matters. Use higher values when you need deeper answers or more room for long outputs.

Practical Recommendations

• use the provider default first
• switch to saver before switching providers just to reduce cost
• use stronger models for repo review, troubleshooting, or migration planning
• use cheaper models for routine checks and repetitive tasks