What problem does it solve?
Pattern-memory solves the problem of improving Council’s routing decisions over time by persistently recording deliberation outcomes, measuring performance (including self-preference bias), and using the resulting data to rebuild routing rules.
Core Features & Use Cases
- Persistent learning without a backend: Maintains a growing, authoritative workspace JSON file (
{workspace}/council-memory.json) that survives Cowork sessions.
- Value-added scoring for routing: Computes and stores
deliberation_value_added to measure whether deliberation beats the best single model for each query category.
- Dynamic smart-router rule rebuilding: Triggers periodic (and manual) router model rebuilds that derive
deliberate vs single_model routing decisions with confidence.
- Bias and performance tracking: Detects potential self-preference bias, tracks model-level aggregates, and records category-wise strengths/weaknesses.
- Optional Obsidian vault mirror: Mirrors key memory data to the user’s Obsidian vault as structured, human-browsable notes to create durable cross-linked knowledge.
Quick Start
Ask the Council plugin to deliberate on a question, then call /council-stats after a few deliberations to see category performance and router status.