Self-Evolution

How Franklin learns your preferences over time.

How It Works

Inspired by NousResearch, Franklin's self-evolution system extracts your preferences after each session. Instead of starting fresh every time, Franklin builds a profile of how you like to work and applies those learnings to future conversations.

What Franklin Learns

Learnings are organized into categories:

Language— preferred language, tone, and communication style
Coding style— formatting preferences, naming conventions, framework choices
Model preferences— which models you prefer for which tasks
Workflow patterns— how you like to break down work, review code, and handle errors

Fully automatic

You don't need to tell Franklin your preferences explicitly. It observes your behavior — corrections you make, models you switch to, patterns in your requests — and extracts learnings automatically.

Confidence Scoring

Each learning has a confidence score. High-confidence learnings (observed repeatedly across multiple sessions) are weighted more heavily. Low-confidence learnings are treated as tentative and may be overridden by newer observations.

Decay & Freshness

Learnings that haven't been reinforced within 30 days gradually decay. This prevents stale preferences from persisting indefinitely — if you change your coding style or switch frameworks, Franklin adapts.

Managing Learnings

View and manage your learned preferences:

bash

# View all current learnings
/learnings

# Clear all learnings and start fresh
/learnings clear

Start fresh anytime

If Franklin's learned preferences feel wrong, use /learnings clear to reset. It will rebuild from scratch in a few sessions.