<2-3 sentences summarising what the project addresses and solves, the target audience, and the benefit.> Cog offers a plain-text cognitive architecture for Claude Code, enabling persistent memory, self-reflection, and foresight. It simplifies AI agent memory management by using observable conventions and the filesystem, allowing models to reason over their knowledge with standard Unix tools and enhancing cross-session context.

How It Works

Cog is a set of plain-text conventions (CLAUDE.md, .claude/commands/*.md) leveraging the filesystem as the interface, eliminating servers or application code. Claude Code interacts with memory using Unix utilities (grep, find, git diff), ensuring observability and maintainability. It features a three-tier memory (hot, warm, glacier) and progressive condensation, transforming raw observations into actionable insights and archives transparently.

Quick Start & Requirements

• Install: git clone https://github.com/marciopuga/cog.
• Run: Open in Claude Code; execute /setup for conversational configuration.
• Prerequisites: Claude Code.
• Permissions: Pre-approved via .claude/settings.json or prompted individually.
• Links: Documentation and Credits referenced in README.

Highlighted Details

• Three-Tier Memory: hot-memory.md (current), domain-specific warm directories, glacier/ (archived).
• Progressive Condensation: Automates observation-to-pattern promotion and archiving.
• Threads: Synthesizes fragmented observations into coherent, long-term narratives.
• Tiered Loading (L0/L1/L2): Optimizes context retrieval via summary, header, and full file scans.
• SSOT: Canonical fact storage with wiki-link references ([[domain/filename]]).
• Pipeline Skills: Workflows (/housekeeping, /reflect, etc.) automate memory maintenance and insights, optionally schedulable.
• MCP Integration: Connects to external tools (Calendar, Gmail, Slack, GitHub) via Model Context Protocol (MCP) for agentic capabilities.

Maintenance & Community

The README does not detail specific contributors, sponsorships, or community channels.

Licensing & Compatibility

• License: MIT.
• Compatibility: Permissive for commercial use and integration with closed-source systems.

Limitations & Caveats

Cog is strictly dependent on the Claude Code environment for execution. It is presented as an experimental system, and its effectiveness relies heavily on the underlying Claude model's ability to interpret and execute the defined conventions and skills.