agnt-gg/agnt

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

AGNT – A Local‑First Agent Operating System

What it is – AGNT is a desktop‑first, self‑hostable platform that bundles an AI‑agent runtime, visual workflow engine, goal manager, memory store, plugin system and a local HTTP API into one “agent operating system”. It is built with Electron + Vue for the UI, an Express backend, and SQLite for persistent data.

Core ideas

Concept What it does When you’d use it
Agents Persistent chat‑style AI workers with memory, tool use, streaming responses and cost accounting. You need judgment, reasoning or a conversational interface that remembers past interactions.
Workflows Drag‑and‑drop DAG canvas with 60+ node types (triggers, branches, checkpoints, nested sub‑workflows). Runs repeatable automations. You want repeatable pipelines – e.g., scrape data, run a model, store results – without writing code.
Goals Long‑running objectives that are planned, executed, evaluated and can be re‑planned, paused or reverted. Projects that span many steps and may need feedback loops, e.g., “launch a product announcement”.
Sub‑agents / delegation Any agent, workflow, tool, plugin, or external service can be invoked as a sub‑task, with full traceability. Parallel or hierarchical work, such as delegating research to one agent while another drafts copy.
Memory & insights SQLite‑backed facts, preferences, corrections, bottleneck detection, and SkillForge that turns execution traces into reusable skills. When you need durable state across runs and want the system to improve over time.
Plugins .agnt packages that add tools, triggers, widgets, hot‑reload support and can be shared via a marketplace. Extending the platform with custom APIs, web‑hooks, or UI components.
Local API HTTP + Socket.IO endpoint (http://localhost:3333/api) exposing agents, workflows, goals, tools, files, plugins, MCP, skills, insights, and executions. Driving AGNT from scripts, other services, or CI pipelines.
Provider freedom Supports 15+ LLM providers (OpenAI, Anthropic, Gemini, Groq, etc.) and custom endpoints. You can switch models or keep everything on‑premise.

Who it’s for

  • Developers who want a UI‑driven way to prototype autonomous AI pipelines without building a backend from scratch.
  • Teams that need local ownership of data, provider keys and execution logs (self‑hosted SaaS alternative).
  • Researchers interested in evaluating, tracing and iterating on agent behaviours (built‑in evals and SkillForge).
  • Power users who want to automate messaging platforms (Telegram, Discord, Slack, email) and combine them with AI.

It is not a simple chatbot or a hosted SaaS automation tool; it requires a local machine or server to run.

Getting started (from the README)

  1. Clone & install
    git clone https://github.com/agnt-gg/agnt.git
cd agnt
npm install
cd frontend && npm install && cd ..
npm start   # launches Electron + backend on http://localhost:3333
  2. Development mode – run the Vue dev server in one terminal (npm run dev in frontend) and the backend in another (npm start).
  3. Docker / headless – pull the image ghcr.io/agnt-gg/agnt:latest and expose port 3333, or build with docker‑compose up -d. Works on x86_64 and ARM64 (Raspberry Pi).
  4. API usage – send JSON to http://localhost:3333/api/agents, .../workflows, etc., and receive Server‑Sent Events / Socket.IO updates for real‑time streaming.

Notable features (as listed)

  • 60+ workflow nodes (triggers, webhooks, Zapier, custom APIs, browser automation via Chromium).
  • Plugin marketplace with 25+ templates.
  • MCP support (multi‑channel protocol) for external tool providers.
  • Self‑hosted first – all data lives in a local SQLite DB and filesystem; you only publish if you choose.
  • Cross‑platform – Windows, macOS, Linux, Docker, Raspberry Pi.
  • SkillForge – automatically extracts reusable instructions from execution traces.
  • Real‑time UX – SSE streams and Socket.IO keep the UI in sync with running agents.

How it fits the AI‑agent ecosystem

Category AGNT’s role
Agent frameworks (LangChain, AutoGen, CrewAI) Provides a full desktop UI, local storage, workflow canvas and plugin system on top of the same underlying agent concepts.
Automation platforms (n8n, Zapier) Adds persistent memory, LLM‑driven reasoning, goal management and self‑improving skills that typical automation tools lack.
Self‑hosted AI workspaces (OpenWebUI, Claude Code) Extends chat with repeatable workflows, sub‑agents, evaluation pipelines and a plugin marketplace.

Quick assessment checklist

  • Do you need local data control? → Yes, AGNT stores everything locally.
  • Do you want a visual workflow builder? → Built‑in DAG canvas.
  • Do you need multi‑model support? → 15+ providers + custom endpoints.
  • Do you only need a simple chatbot? → Probably overkill; consider a lighter UI.

Bottom line

AGNT is a genuine, open‑source “agent operating system” that combines the capabilities of an LLM‑agent runtime, a low‑code workflow engine, a goal‑oriented planner and a plugin marketplace into a single, self‑hostable desktop application. It is aimed at developers and power users who want durable, inspectable AI automation that they can run and extend on their own hardware.

All information above is taken directly from the repository’s README; no external assumptions have been added.

Related

  • Project
    hAcKlyc/MyAgents

    MyAgents is an open‑source, local‑first desktop workbench for creating and orchestrating AI agents. It supports multiple LLM runtimes, a task scheduler, IM channel integration, and a private device‑to‑device network (AgentNet) for multi‑computer workflows, all under an AGPL‑v3 license.

  • Project
    eggent-ai/eggent

    Eggent is a self‑hosted, browser‑based AI workspace that lets you run LLM‑backed agents per project, store files and memory locally, chain agents with pipelines, and integrate via a simple HTTP API or Telegram bot. It ships as a Docker image, uses local storage under `data/`, and supports on‑device voice transcription with whisper.cpp.

  • Project
    androoAGI/starnet

    StarNet is a local‑first desktop app that lets you build and run multiple AI agents in a pixel‑art space‑station UI. Each room represents a bounded team with its own memory, permissions and budget. Agents use real LLM providers (OpenAI, Anthropic, Google, OpenRouter, or a local Ollama model) and can call tools, send messages to chat platforms, run scheduled recipes, and output real files. The visual layout is the live workflow, and all state (transcripts, spend logs, memory) is persisted on disk with secrets kept in the OS keychain.

  • Project
    AtomicBot-ai/atomic-agent

    A local-first, open-source AI agent that runs a control loop on your machine, can operate a browser, edit files, run approved commands, and keeps persistent memory — designed to work efficiently with small quantized models via llama.cpp.

  • Project
    mrslimslim/gpt-image-canvas

    A local-first AI image workspace that combines an infinite canvas with agentic planning for multi-step prompt-to-image generation.