agnt-gg/agnt

The local-first operating system for building, running, and improving AI agents, workflows, and autonomous goals.

What it solves

AI agents often fail because they are confined to disposable chat sessions. AGNT provides a durable, local-first runtime environment that allows agents to maintain state, use trusted tools, execute repeatable workflows, and pursue long-running goals beyond a single conversation.

How it works

AGNT operates as a local-first "agent operating system" consisting of a desktop application and a self-hostable backend. It utilizes three primary execution modes:

  • Agents: For reasoning, tool use, and persistent chat with memory.
  • Workflows: A visual DAG engine for repeatable automation using triggers and nodes.
  • Goals: A system for planning and executing long-term objectives through task decomposition and evaluation.

These modes are interconnected; for example, a workflow can call an agent, and a goal can use execution traces to improve skills. The system is built with Electron, Vue.js, Express, and SQLite for local data control.

Who it’s for

It is designed for users who need repeatable AI automations, long-running objectives, delegated work across multiple agents and tools, and full ownership of their data and provider settings through local or self-hosted deployments.

Highlights

  • Local-First Architecture: Data, traces, and settings are stored locally in SQLite and the filesystem.
  • Subagent Delegation: Native support for delegating tasks to other agents, workflows, async tools, and MCP servers.
  • SkillForge: A feature that converts execution traces into reusable skills.
  • Extensive Integration: Supports 15+ AI providers, 60+ built-in tools, and MCP (Model Context Protocol) support.
  • Flexible Deployment: Available as a desktop app, Docker containers (Full and Lite), or a headless server for low-power devices like Raspberry Pi.