Akm, developed by Itlackey, is a command-line server tool designed to manage AI agent knowledge, memories, skills, and more. Built primarily with TypeScript, it acts as a package manager for AI agent capabilities. Akm integrates with various sources including local directories, Git repositories, npm packages, and websites, allowing agents to discover and incorporate new functionalities. It supports both npm package and prebuilt binary distributions, catering to different deployment needs. This free tool helps coding agents and teams organize and utilize their collective intelligence effectively.

What does Akm connect to?

Akm connects to a variety of external and internal systems to gather and manage agent knowledge. It integrates with local directories, Git repositories, npm packages, and websites. The tool also supports user-configured model endpoints and registries. Specifically, Akm can connect with GitHub repositories, functioning as both Git sources and OpenViking providers. Its network usage is limited to explicitly configured remote sources and endpoints, ensuring a local-first operational model for its core index and state.

Core features and tools

Akm provides a set of callable tools for managing AI agent capabilities. Key commands include akm search for discovery, akm add and akm clone for integration, and akm show for inspection. It also offers akm list, akm update, akm remove, and akm upgrade for lifecycle management. Specialized subcommands exist for knowledge organization, such as akm wiki create, and for workflow orchestration, like akm workflow next. The tool also includes akm vault for secret management and akm save for syncing changes.

Who is it for?

Akm is designed for "every coding agent" and "teams" that need to manage AI agent capabilities. It serves "any shell-capable coding agent" looking to organize and access skills, memories, and knowledge efficiently. The tool supports collaborative knowledge management, making it suitable for groups aiming to support the improvement of their agent libraries through reviewable proposals. Akm helps agents discover, integrate, and execute capabilities from various sources. Skip this if your AI agents cannot execute shell commands.

How to get started

The server authenticates using an API key passed as an environment variable for remote context servers. Akm supports HTTP and stdio as transport protocols for communication.

Use cases

• Capability Discovery and Integration: An AI agent can use "akm search" to find relevant skills or scripts, "akm show" to inspect them, and then "akm add" or "akm clone" to integrate these capabilities into its operational context.
• Execution of Durable Workflows: Agents can run multi-step, persisted workflows with features like dispatch, gates, retries, budgets, and resume capabilities, using commands such as "akm workflow next" to deterministically progress through procedures.
• Collaborative Knowledge Management: Teams can use Akm to capture memories, import documentation, manage wikis, and support the improvement of the agent library through reviewable proposals, utilizing commands like "akm wiki create" and "akm save" for syncing changes to remote Git repositories.

Akm offers a specialized command-line interface for managing AI agent knowledge and capabilities. Its local-first design, combined with support for diverse data sources and AI clients, makes it a useful tool for agent development. The system provides tools for discovery, integration, and execution of agent workflows, alongside collaborative knowledge management. Akm is a free, TypeScript-based solution for organizing AI agent intelligence.