Yome Agent provides an open-source substrate, the "Empower Kernel," designed to integrate Large Language Models (LLMs) with users' existing native applications and devices. It aims to automate the tedious task of context and tool routing, which currently consumes significant user time, by enabling AI to directly interact with the tools users already possess. This approach targets power users and engineers seeking a more integrated and proactive AI runtime that operates ambiently and reacts to real-world events.

How It Works

Yome Agent operates on three interlocking engines: the Agentic Contextual Engine for always-on, memory-rich context; the Agentic Native Skill engine for invoking local applications; and the Agentic Bash Kernel for a unified command interface. Its core innovation lies in waking up existing native apps (like PowerPoint, Excel, Mail) via "Native Skills," which declare required OS resources and require explicit user permission grants, creating a sandboxed execution environment. The Bash Kernel acts as a dispatcher, routing commands either to the native shell or to installed skills, enabling seamless composition between shell commands and application actions.

Quick Start & Requirements

Install via npm: npm install -g @poping/yome. Configure with API keys, base URL, and model: yome --key sk-... --base-url https://your.endpoint --model your-model. Run interactively: yome or for one-shot commands: yome "summarise package.json". Example skill installation: yome skill install github:Whopus/yome-skill-ppt. Requires Node.js (implied by npm). API keys and a configured LLM endpoint are necessary for full functionality.

Highlighted Details

• Native Skills & Capability Model: Invokes local applications via platform-specific scripting (AppleScript, Win32, DBus) with a robust capability model requiring explicit user permission grants for actions like filesystem access or shell execution.
• Unified Bash Kernel: Acts as a single interface for real shell commands, skill verbs, and shell-skill composition, simplifying LLM tool interaction.
• Batch Mode Performance: Optimizes sequential native app actions (e.g., PowerPoint edits) by fusing multiple steps into single calls, achieving up to a 6x speedup compared to separate invocations.
• Skills Marketplace: A decentralized hub for discovering and installing community-contributed skills, with core skills licensed under Apache 2.0, while community skills retain their own licenses.

Maintenance & Community

The project maintains official skills (e.g., @yome/ppt, @yome/xl) with stable or beta status. Community-contributed skills are hosted as plain Git repositories on a public hub, avoiding vendor lock-in. Roadmap details for future features like Daemon and Oncall are partially outlined, with Daemon currently in experimental status. No direct community links (Discord, Slack) are provided.

Licensing & Compatibility

The core Yome Agent and its official skills are licensed under the Apache License 2.0, permitting commercial use and modification. Community-contributed skills published to the hub are subject to their respective author-owned licenses.

Limitations & Caveats

Key features like the OS-level Daemon and event-driven Oncall capabilities are currently in experimental or "next-up" development stages. Some native skills are marked as beta or alpha, indicating ongoing development and potential instability. The system relies on users having the target native applications already installed on their machines.