How it works
From imported conversation to reviewable project memory.
The model is not the database. The Brain owns the project record, assembles context for the model, and keeps every important write behind a review boundary.
- 01
Capture
Conversations, pasted transcripts, shared links, repository docs, notes, and mail enter the inbox with source metadata.
- 02
Classify
The engine proposes projects, tasks, decisions, resources, blockers, and relationships. Unsupported claims stay marked.
- 03
Assemble
Before a model answers, relevant workspace context is discovered, ranked, trimmed to a budget, and sent with the prompt.
- 04
Review
AI output becomes proposed records. The user approves, edits, rejects, archives, or deletes them.
- 05
Remember
Approved records become durable project memory and are available to future Codex, Claude, ChatGPT, and local model workflows.