AI coding agents often produce valuable output, but their underlying processes can be difficult to inspect or repeat. Cognetivy addresses this by providing an open-source state layer that transforms chaotic agent sessions into structured, traceable workflows. It offers a local workspace for managing runs, events, and collections, making AI-assisted coding more auditable and repeatable. This tool is particularly beneficial for developers building reproducible AI coding workflows, researchers executing structured tasks, and teams requiring traceability for agent outputs.

How It Works

Cognetivy functions as an operational layer for AI coding agents, capturing their activities within a local .cognetivy/ workspace. It allows for the definition of explicit workflows, detailed tracking of individual runs and events, and the organization of reasoning artifacts into structured collections. The system operates without embedding LLMs directly, instead managing the data and tools used by agents through Skills and MCP-compatible clients. This design facilitates inspection and repetition of agent processes, turning ephemeral chat histories into persistent, auditable operational layers.

Quick Start & Requirements

• Requirements: Node.js version 18 or higher, a project directory (or empty folder) to serve as the workspace, and a compatible AI coding agent (e.g., Claude Code, Cursor, OpenClaw).
• Installation: For a one-time setup and installer, use npx cognetivy. For global use and MCP integration, install via npm install -g cognetivy.
• Setup: Running npx cognetivy launches a terminal installer to select an agent, create the .cognetivy/ workspace, and install necessary skills. Cognetivy Studio then opens in the browser. MCP integration involves configuring Cognetivy as a tool server within compatible editors.
• Links: Cognetivy Website, Workflow Templates.

Highlighted Details

• A persistent local .cognetivy/ workspace stores workflows, runs, events, and collections.
• Enables the definition of explicit, multi-node workflows for structured agent execution.
• Integrates with AI coding agents via Skills and MCP, making tools available within editor chats.
• Includes built-in workflow templates for common use cases, such as bug triage and fixing.

Maintenance & Community

The project is actively maintained, with contributions welcomed through issues and pull requests. A Code of Conduct is established. The README does not specify community channels like Discord or Slack, nor does it highlight specific contributors.

Licensing & Compatibility

The project is licensed under the MIT license. This license generally permits broad usage, including commercial applications and linking within closed-source projects without significant restrictions.

Limitations & Caveats

The provided README does not explicitly list project limitations, known bugs, or indicate an alpha status. It emphasizes active maintenance and an open contribution model.