Skip to content

Effort Command /effort

Overview

The /effort slash command sets the reasoning effort level used by the active LLM. A single knob covers every supported provider: LiteLLM maps the level to each provider's native dialect — OpenAI reasoning_effort, Anthropic thinking.budget_tokens, Gemini thinking_config.thinking_budget, DeepSeek / Kimi reasoning, etc.

The setting can be persisted globally (in agent.yml) or scoped to the current project (in .datus/config.yml); the project-level value always wins at runtime.

Effort Levels

Level Meaning
off Disable reasoning (no thinking)
minimal Minimal effort (fast; gpt-5 family)
low Low effort
medium Medium effort (balanced)
high High effort (deep reasoning)

If the active model does not support reasoning, the level is silently ignored. /effort status shows whether the active model can actually consume the hint (queried via litellm.supports_reasoning).

Basic Usage

Interactive picker

Type /effort with no arguments to open the TUI. The picker is a two-step flow:

  1. Pick an effort level
  2. Pick the persistence scope (project vs global)
/effort

Direct shortcuts

# Pick level, then choose scope interactively
/effort high
/effort minimal

# Persist to project (.datus/config.yml)
/effort high --project

# Persist to global (agent.yml)
/effort high --global

# Disable reasoning
/effort off

# Remove the project-level override (fall back to global / model defaults)
/effort --clear

# Show the effective level and where it came from
/effort status

Resolution Order

When the agent starts a turn, the effective effort is resolved in this order (first match wins):

  1. .datus/config.yml → reasoning_effort: (project override)
  2. agent.yml → top-level reasoning_effort: (global default)
  3. agent.models.<active>.reasoning_effort (model-level config)
  4. agent.models.<active>.enable_thinking: true → treated as medium
  5. Otherwise: not set (model-level defaults apply)

/effort status prints the resolved level alongside its source label (project / global / model / not set).

Examples

# One-off: switch to high effort for the current project
/effort high --project

# Make "low" the default for every project on this machine
/effort low --global

# Quickly turn reasoning off for the current project
/effort off --project

# Reset and let the model defaults apply again
/effort --clear

# Inspect what's currently effective
/effort status

See also: /model for switching the active LLM.