huytieu/COG-second-brain

What it solves

COG is an agentic "second brain" designed to move beyond static note-taking. It solves the problem of manual knowledge management by using AI agents to automatically classify braindumps, synthesize insights, and maintain a structured knowledge base from scattered markdown files without requiring a centralized database or vendor lock-in.

How it works

COG transforms a local folder of markdown files (compatible with Obsidian) into an intelligent system. It provides a set of "skills" (documented playbooks) that AI agents (such as Claude Code, Cursor, or Gemini CLI) can execute. The system uses a worker-agent architecture where specialized workers handle data-heavy I/O tasks (using cheaper models like Sonnet) while a lead agent handles high-level reasoning (using models like Opus).

To ensure accuracy, it employs a "Verification Harness" where read-only verifiers independently observe artifacts to validate work against acceptance criteria, preventing the AI from "grading its own homework."

Who it’s for

• Knowledge Workers: People who want an automated, privacy-first personal knowledge management system.
• Product Managers: Those needing a full lifecycle workflow from research and PRDs to user stories and release notes.
• Engineering Leads: Users who want to cross-reference team intelligence from GitHub, Slack, and Linear into daily briefs.

Highlights

• Self-Evolving: Automatically learns user patterns and consolidates scattered notes into frameworks.
• Multi-Agent Architecture: Uses specialized worker agents for data collection, research, and publishing.
• Verification-First: Implements a V-model closed loop with independent verifiers to ensure evidence-based results.
• Privacy-First: Operates entirely on local markdown files with optional Git/iCloud sync.
• Broad Agent Support: Native surfaces for Claude Code, Antigravity, Cursor, Kiro, and Gemini CLI.