• Claude Code shipped an in-app browser running in sandbox mode — no history saved, no logins stored
• Google Cloud Run sandboxes hit public preview — zero-trust, deny-by-default network egress, millisecond startup
• Northflank formalized the "AI SDLC" — with sandboxed agent execution as the first lifecycle stage

Everyone is converging on the same realization: AI agents don't just need to generate code. They need a real place to work.

We call this the "desk moment." Human developers have desks — computers with filesystems, terminals, browsers, git, and colleagues. AI agents deserve the same. Not a chat window. A real workspace.

AinoWork is powered by Hermes Agent — Nous Research's open-source LLM agent framework with self-improving agent loops. AinoWork adds the UI, the sandbox, and the multi-user layer on top.

The Gap

What's missing is a workbench — where AI agents have the same tools as human developers.

AinoWork is an open-source, self-hostable team AI workbench. It fills that gap, and it's designed for both individual developers and enterprise teams.

How We Designed It

1. Give AI a Real Desk (Sandbox-First)

Every agent session spawns a fresh Docker container with a real Linux filesystem, process isolation, and network access. It can npm install, pytest, git push — and if it breaks something, the blast radius is one container. The sandbox isn't a restriction. It's what makes bold experimentation safe.

2. Full Toolchain, Not Just Chat

The agent's "desk" comes equipped with: file browser, WebSocket terminal, VNC-controlled Chromium, and git (clone, branch, diff, commit, push). Plus Hermes Agent for self-improving agent loops — the agent learns from experience.

3. Enterprise-Ready, Open by Default

Self-hosted. 200+ LLM models. Data stays on your infrastructure — deploy on-premises, in your VPC, or air-gapped. Bring your own API keys — no vendor lock-in. Skills become team assets — reusable across the organization, retained when people leave.

Why Not Just a Generic Cloud Agent?

Personal cloud AI tools make one person faster. AinoWork gives your team a shared, self-hosted AI workspace.

What This Means

If you're an individual developer: an AI pair programmer that actually runs code, tests web apps, and commits changes — all in one browser tab.

If you're a team: a shared AI workspace with RBAC, per-user model configs, and isolated sandboxes per session. Data stays in your boundary, models are controlled with your own keys, and skills survive turnover.

One Command to Start

curl -fsSL https://raw.githubusercontent.com/oinone/ainowork/refs/heads/master/docker/ainowork-starter.zip -o ainowork-starter.zip
unzip ainowork-starter.zip && cd ainowork-starter
bash start-mac.sh

Open https://localhost:3000, configure your LLM API key, and you're in.

Current status: v1.0.0 — 23 pluggable SPI modules, 200+ LLM models, full i18n (EN/ZH), actively developed with community input.

FAQ

Q: How is this different from Cursor or ChatGPT? A: Different category, complementary relationship. Cursor is an AI-enhanced code editor — great for writing and refactoring code. ChatGPT is a powerful AI assistant for analysis and problem-solving. AinoWork is an AI developer workbench powered by Hermes Agent — it provides Docker sandboxes where AI agents can execute code, test web apps, browse, and commit changes. Think of it as an enterprise-grade interface for Hermes Agent: workbench, not editor or chatbot.

Q: Can I use my own LLM API keys? A: Yes. AinoWork supports 200+ models across OpenAI, Anthropic, and any OpenAI-compatible endpoint. Bring your own key.

Q: Is this production-ready for enterprise teams? A: v1.0.0 ships with RBAC, bring-your-own-key, and private deployment options. It's being actively developed with community input.

AinoWork on GitHub →