AGNT
Get more from the AI you already pay for
About AGNT
AGNT is a local-first operating system for AI agents that runs on your own machine rather than in someone else's cloud. Instead of asking you to buy yet another AI subscription, it connects to the ones you already pay for, alongside your API keys and local models, and gives that access a place to do real work. The problem it addresses is familiar to anyone who has accumulated a ChatGPT plan, a Claude plan and a coding agent: the intelligence is already paid for, but it lives in separate chat tabs that produce advice rather than finished output. AGNT supplies the workspace, the tools and the record that turn a conversation into a completed job.
The runtime centres on Annie, a local assistant that builds and coordinates agents, workflows, tools and skills. Annie keeps persistent memory across conversations, reaches roughly 100 built-in tools covering web search, scraping, file operations, code execution, media generation, email, spreadsheets, Git and shell, and can run work asynchronously in the background on intervals or after a delay. AGNT can also delegate to external execution engines including Grok CLI, Codex, Claude Code, Hermes and OpenCode. The repository at github.com/agnt-gg/agnt carries roughly 540 stars and is under active development, though it is worth noting the licence is a custom one rather than an OSI-approved open source licence.
You install AGNT as a native desktop application on macOS, Windows or Linux, run it headless with a single Docker command, or host the full stack on your own servers. On first run you connect the model access you already have, whether that is a supported AI subscription, an API key, or a local runtime such as Ollama or LM Studio.
From there you describe the outcome you want rather than designing an agent up front. AGNT plans the work, requests only the tools and authority the operation actually needs, and executes. A run that hits a blocker stays blocked rather than being reported as complete. When it finishes you get a receipt: the sources consulted, the actions taken, the approvals given, the cost incurred and any unresolved failure. Repeatable work can then be promoted into a visual workflow that connects triggers, tools and code.
- •Annie, the local agent - A persistent assistant with memory that survives across conversations, roughly 100 real tools and 30 or more skills that activate contextually for tasks such as research synthesis, codebase analysis and media pipelines.
- •Bring your own AI - Connects supported subscriptions from ChatGPT, Claude, Google AI, Grok and others alongside 20 or more model providers and local runtimes, so you choose a model per task instead of being locked to one vendor.
- •Visual workflow builder - Connect triggers, tools and code on a canvas, inspect the configuration next to the graph, and give repeatable work a repeatable structure with approval gates where judgement is needed.
- •Receipts and cost visibility - Every run records its sources, actions, approvals, failures and cost, keeping metered API spend separate from subscription fees rather than presenting one opaque total.
- •50+ native integrations - 119 tool nodes spanning GitHub, Slack, Stripe, Postgres, Gmail, Notion and Google Workspace, plus a Zapier bridge to thousands more apps and a client for any MCP server.
- •Runs where you run - Native desktop builds with an OS keychain vault, a one-command Docker deployment for a NAS or VPS, full self-hosting, or AGNT-operated cloud when you would rather not run infrastructure.
Developers and technical operators who already pay for several AI services and want more from them than a chat window will get the most out of this. It suits research work such as decision briefs and competitor monitoring, operations work like KPI reporting and reconciliation, customer work including lead qualification and proposals, and engineering tasks from issue investigation through release. Teams with data they cannot send to a third-party cloud will find the local-first and self-hosted deployment options particularly relevant, since the runtime, the vault and the receipts all stay on infrastructure they control. It is a poor fit for anyone wanting a zero-setup consumer chatbot, and for work where every step is known in advance, conventional deterministic automation remains the simpler answer.
Pricing
$0 - $99/mo
- Community Core$0/mo
- Personal Cloud$29/mo
- Business Cloud$99/mo
From the vendor pricing page, 2026-09-18
