agent

Engage with an autonomous agent which automatically generates and uses classifiers or steering vectors according to your query. This agent studies your directive, makes essential parts and performs quality assurance.

Basic Usage

python -m wisent agent "YOUR PROMPT" [OPTIONS]

Examples

Basic Agent Interaction

python -m wisent agent \
  "Generate a response that is more helpful and detailed" \
  --model meta-llama/Llama-3.1-8B-Instruct

With Steering Mode

python -m wisent agent \
  "Respond like a friendly expert teacher" \
  --model meta-llama/Llama-3.1-8B-Instruct \
  --steering-mode \
  --steering-method CAA \
  --steering-strength 1.5

With Quality Control

python -m wisent agent \
  "Answer questions more honestly" \
  --model meta-llama/Llama-3.1-8B-Instruct \
  --enable-quality-control \
  --max-quality-attempts 5 \
  --quality-threshold 0.3 \
  --verbose

With Time Budget

python -m wisent agent \
  "Be more concise in responses" \
  --model meta-llama/Llama-3.1-8B-Instruct \
  --time-budget 5.0 \
  --max-classifiers 3

Arguments

Required

Argument	Description
prompt	Prompt to send to the autonomous agent

Model Configuration

Argument	Default	Description
--model	Llama-3.1-8B-Instruct	Model to use
--layer	from config	Layer to use (overrides parameter file)

Agent Settings

Argument	Default	Description
--quality-threshold	0.3	Quality threshold for classifiers
--time-budget	10.0	Time budget in minutes for creating classifiers
--max-attempts	3	Maximum improvement attempts
--max-classifiers	unlimited	Maximum classifiers to use

Steering Options

Argument	Default	Description
--steering-mode	false	Enable steering mode
--steering-method	CAA	Steering method to use
--steering-strength	1.0	Steering strength

Normalization

Argument	Default	Description
--normalize-mode	false	Enable normalization of steering vectors
--normalization-method	none	Normalization method (none, l2_unit, l2_norm, max_norm)
--target-norm	None	Target norm for steering vectors

Quality Control

Argument	Default	Description
--enable-quality-control	true	Enable quality control system
--max-quality-attempts	5	Maximum attempts to achieve acceptable quality
--show-parameter-reasoning	false	Display model's reasoning for parameter choices

How the Agent Works

Analyze Prompt - Agent analyzes your prompt to understand desired behavior
Create Components - Automatically creates classifiers or steering vectors
Quality Check - Validates that components meet quality thresholds
Apply & Generate - Applies components and generates the response
Iterate if Needed - Refines components if quality is below threshold

Related Commands

tasks - Run evaluation tasks
synthetic - End-to-end pipeline
evaluate - Evaluate single prompts

Stay in the loop. Never miss out.

Subscribe to our newsletter and unlock Wisent insights.

Contact Careers Privacy Policy Terms of Service