LLM_wiki_personal_brain

Local-first Personal Brain OS with a durable wiki middle layer and a Step3 asset-production pipeline.

Step4 Brain Workbench

This repo now also includes a dedicated frontend workbench in brain-workbench/.

The frontend is a thin Vue workbench over the Personal Brain OS backend, with pages for:

• ask workspace
• writeback review
• asset candidates
• profile workspace
• eval reports
• wiki explorer

Frontend API binding uses the Python apps/api/server.py layer and falls back to local mock payloads when an endpoint is unavailable.

Layer Model

The repository keeps four strict layers:

• raw/ preserves source truth and provenance.
• wiki/ is the primary human-readable compiled knowledge layer.
• ontology/ and memory/ store structured assets derived from wiki-backed evidence.
• src/personal_brain/agent/ and skills/ provide the runtime and candidate skill surfaces.

The system now optimizes for durable asset production, not just grounded answers.

What Step3 Adds

Step3 upgrades the existing Step2 query engine into a true asset-production system:

• structured writeback routing plus merge-safe wiki updates
• method profile and template selection that shape structure more than tone
• ontology candidate extraction from wiki pages
• skill candidate generation from repeated high-value sessions
• evaluation harness for “worth sinking” quality, not just code correctness

Writeback Pipeline

writeback is now a routed flow: