lukilabs

craft-agents-oss

// summary

Craft Agents is an open-source, agent-native desktop application designed to provide an intuitive and document-centric interface for interacting with powerful AI models. It supports seamless integration with various APIs, MCP servers, and local filesystems, allowing users to connect services without complex configuration files. The platform features a robust session management system, multi-provider LLM support, and a headless server mode for advanced remote workflows.

// technical analysis

Craft Agents is an agent-native desktop application designed to provide a fluid, document-centric workflow for interacting with AI agents. By integrating both the Claude Agent SDK and the Pi SDK, it enables seamless multitasking and connectivity to diverse APIs, MCP servers, and local filesystems without requiring complex configuration files. The project prioritizes a highly customizable, non-CLI user experience, allowing users to manage agent sessions, skills, and automations through an intuitive UI while maintaining the flexibility to run as a headless server for remote compute tasks.

// key highlights

Connects to any service or API, including Slack, Gmail, and Linear, by simply describing the intent to the agent.

Supports both local and remote MCP servers, allowing for seamless integration of local scripts, binaries, and databases.

Features a multi-session inbox with customizable workflow statuses like Todo, In Progress, and Done to manage agent tasks effectively.

Provides a three-level permission system (Explore, Ask to Edit, Auto) to control agent autonomy and ensure secure operations.

Offers a headless server mode that allows the desktop app to act as a thin client, enabling remote execution of compute-heavy tasks.

Includes a robust automation engine that triggers actions based on events such as label changes, schedules, or tool usage.

// use cases

Connecting to external services like Slack, Gmail, and Linear via MCP servers or REST APIs without manual configuration.

Managing complex AI-driven workflows through a multi-session inbox with customizable status tracking and permission modes.

Running compute-intensive agent tasks on remote headless servers while using a local desktop app as a thin client.

// getting started

To begin, download and install the application using the provided one-line install script for your OS or build it from source using Bun. Once launched, configure your preferred LLM provider (such as Anthropic, OpenAI, or Google) in the settings, create a workspace, and start interacting with agents by connecting your desired sources.