How pref0 learns formatting and output preferences

How output is formatted matters as much as the content itself. pref0 learns whether users prefer markdown, JSON, bullet points, tables, or other formats and applies these preferences consistently.

What pref0 extracts

Examples of formatting preferences that pref0 detects and scores from conversations.

"Give me a markdown table"

output_format: markdown_table0.40
Explicit instruction

"JSON, not YAML"

config_format: json0.70
Explicit correction

"Always include code comments"

code_comments: always0.40
Explicit instruction

"Use 2-space indentation"

indentation: 2_spaces0.70
Explicit instruction

"Wrap at 80 characters"

line_width: 800.40
Explicit instruction

"Include type annotations"

type_annotations: always0.40
Explicit instruction

Confidence progression

See how a single preference compounds in confidence across multiple conversations.

Session 1

"Use 2-space indentation"

0.70
Session 2

User formats shared code with 2 spaces

0.80
Session 4

"Indent with 2 spaces, not 4"

0.95
Session 5

Consistent 2-space usage confirmed

1.00

Related preference types

See preference learning in action

Your users are already teaching your agent what they want. pref0 makes sure the lesson sticks.