AGNT
A local-first agent operating system for building, running, and evolving AI workflows, agents, and goals on your own infrastructure.
At a Glance
Free forever plan with core features and community support, includes usage limits.
Engagement
Available On
Alternatives
Listed Jun 2026
About AGNT
AGNT is a local-first agent operating system built by Nathan Wilbanks, available as a desktop app and self-hostable backend. It brings agents, workflows, goals, memory, plugins, evaluations, and APIs into one unified local workspace, running on Windows, macOS, and Linux. The project is hosted on GitHub under a custom license and is currently at version 0.5.17.
What It Is
AGNT positions itself as a full agent operating system rather than a single-purpose automation tool or chatbot. It combines three core execution modes — Agents (for judgment and reasoning), Workflows (for repeatable automation via a visual DAG engine), and Goals (for long-running autonomous objectives) — into one local runtime. The backend runs on Express with SQLite storage, the frontend is built in Vue 3, and the desktop shell uses Electron. All data stays on your infrastructure by default.
Architecture and Runtime Model
AGNT's architecture is a layered stack: an Electron shell wraps a Vue 3 frontend, which communicates over HTTP, SSE, and WebSocket with a local Express backend on port 3333. The backend hosts the agent runtime, workflow engine, goal engine, plugin loader, MCP client, provider auth dispatcher, and SkillForge system, all backed by SQLite and the local filesystem.
- Agents provide persistent chat, memory, tools, skills, streaming, and token/cost accounting.
- Workflows offer a visual DAG engine with 60+ built-in nodes, triggers, branches, checkpoints, and nesting.
- Goals support planning, task decomposition, execution, evaluation, re-planning, pause/resume/revert, and golden standards.
- SkillForge converts execution traces into reusable, improvable skills.
- Plugins are
.agntpackages with tools, triggers, widgets, hot reload, and marketplace distribution.
Supported AI Providers and Integrations
AGNT uses a pluggable provider layer supporting 15+ AI providers, including OpenAI, Anthropic, Google Gemini, Grok, Groq, Cerebras, DeepSeek, OpenRouter, Together AI, Kimi, MiniMax, Z.AI, Claude Code CLI, OpenAI Codex CLI, Gemini CLI, and custom OpenAI-compatible endpoints. Integration reach extends to 10,000+ services via MCP Protocol, plus Zapier/n8n integrations and 120+ native connectors. Built-in workflow triggers and actions cover Discord, Slack, Gmail, GitHub, Notion, Stripe, Google Sheets, Google Drive, Dropbox, Twitter/X, YouTube, Firecrawl, Unsplash, and more.
Deployment and Setup Paths
AGNT supports multiple deployment modes to fit different infrastructure needs:
- Desktop binaries: Pre-built installers for Windows 10/11, macOS 12+ (Apple Silicon and Intel), and Linux (Debian/Ubuntu .deb, AppImage, Red Hat/Fedora .rpm).
- Source install:
git clone+npm install+npm start; backend runs on port 3333. - Docker: Full image (~1.5 GB, includes Chromium for browser automation) and Lite image (~715 MB, without browser automation), pulled from GHCR.
- Headless/VPS/Raspberry Pi: Lite Docker image recommended for $5 VPS or Raspberry Pi 4/5 (64-bit OS required); Node.js 18+ and npm 9+ are prerequisites.
Update: Version 0.5.17
The latest release is v0.5.17, published on May 29, 2026. The GitHub repository was created in March 2025 and has accumulated 352 stars and 50 forks as of the latest data. The project uses tag-driven releases with CI publishing multi-arch Docker images to GHCR. The README describes an active release cadence with patch, minor, and major release scripts. The product direction signal is toward a comprehensive local agent OS that can self-improve through SkillForge and support delegation graphs across agents, workflows, tools, plugins, and MCP servers.
Who It Is For
The homepage states AGNT is designed for developers, DevOps/Ops teams, and founders/CTOs who need local-first automation with full data sovereignty. The README identifies good fits as single users, small teams of 2–10, developers wanting an inspectable local agent backend, and automation builders needing visual workflows plus AI judgment. It explicitly notes it is not designed for public multi-tenant SaaS, large enterprises with 50+ concurrent users, or zero-trust tenant isolation requirements.
Community Discussions
Be the first to start a conversation about AGNT
Share your experience with AGNT, ask questions, or help others learn from your insights.
Pricing
Community Core
Free forever plan with core features and community support, includes usage limits.
- Full core features
- Community support
- Limited workflows
- Basic automations
- Core integrations
Personal Pro
Unlimited workflows and automations for individual users with private cloud sync and personal API access.
- Full core features
- Community Pro support
- Private cloud sync
- Webhook / email access (sync every 15 min)
- Personal API access
- Unlimited workflows
- Unlimited automations
- All integrations
- No usage limits
- No multi-user
- No white-label
- No SLA
Business Pro
Unlimited workflows and automations for teams up to 10 users with real-time sync and business API access.
- Full core features
- Business Pro support
- Private cloud sync
- Webhook / email access (realtime sync)
- Business API access
- Multi-user (up to 10)
- Unlimited workflows
- Unlimited automations
- All integrations
- No usage limits
- No white-label
- No SLA
Enterprise
Unlimited users, white-label, custom SLA, and dedicated 24/7 support for large organizations.
- Full core features
- 24/7 dedicated support
- Private cloud sync
- Webhook / email access (realtime sync)
- Unlimited API access
- Unlimited users
- White-label
- Custom SLA
- Unlimited workflows
- Unlimited automations
- All integrations
- No usage limits
Capabilities
Key Features
- Local-first architecture with 100% data sovereignty
- Visual workflow canvas with 60+ built-in nodes
- Persistent AI agents with memory, tools, and skills
- Goal engine with planning, evaluation, and re-planning
- SkillForge: convert execution traces into reusable skills
- Plugin system (.agnt packages) with hot reload and marketplace
- 15+ AI provider integrations (OpenAI, Anthropic, Gemini, etc.)
- MCP (Model Context Protocol) client support
- 10,000+ integrations via MCP/API/Plugin connectors
- AI-assisted workflow generation from plain English
- Multi-agent delegation and subagent orchestration
- Real-time monitoring, logging, and debugging
- REST API + WebSocket + SSE for programmatic control
- Docker support (Full and Lite variants)
- Desktop binaries for Windows, macOS, and Linux
- Headless/VPS/Raspberry Pi deployment mode
- Browser automation (Chromium/Puppeteer/Playwright in Full image)
- TypeScript/JavaScript SDK with full API coverage
- Built-in evals, experiments, and insights system
- Air-gap compatible with zero external dependencies
