holaOS is an open-source agentic desktop workspace where AI agents work beside apps, browser sessions, local files, chat conversations, and connected services.
It can run Claude Code, Codex, or the built-in holaOS agent while you inspect or interact with the apps and files they are working on.
The workspace keeps project instructions, skills, integrations, MCP connections, and memory available across sessions. You can switch agents or models without rebuilding the project context.
Key Features
- Opens interactive apps beside the active agent and lets you take manual control when needed.
- Connects Slack, Feishu, DingTalk, WeChat, Gmail, Notion, GitHub, Linear, and other services.
- Runs Claude Code, Codex, and the built-in holaOS agent inside one persistent workspace.
- Works with Holaboss-managed models, your provider keys, OpenRouter, and local models.
- Stores workspace files, memory, embeddings, session history, and runtime state on your machine.
- Packages recurring instructions as reusable skills.
- Connects local or remote MCP servers to workspace apps and agent actions.
- Lets agents work with a signed-in browser, local files, and connected services.
- Creates
.docx,.xlsx, and.pptxdeliverables. - Installs community apps, skills, and MCP resources from holahub.
How holaOS Works
Each holaOS workspace holds the resources an agent needs for ongoing work: files, standing instructions, skills, apps, MCP configuration, runtime state, and durable memory. The active agent receives the capabilities available for the task while the workspace keeps the project context between sessions.
HolaApps open web apps and custom applications beside the agent. You can interact with the app directly while the agent uses its available actions. Chat connections and service integrations bring conversations and external data into the workspace, while MCP servers expose local or remote actions the agent can call.
| Role | |
|---|---|
| Agents | Claude Code, Codex, or the built-in holaOS agent |
| Apps | Interactive web apps, custom apps, and app actions |
| Skills | Reusable instructions for recurring tasks |
| MCP | Local or remote MCP servers and available actions |
| Integrations | Connected services such as Gmail, Notion, Slack, GitHub, and Linear |
| Memory | Workspace knowledge, preferences, identity data, and session continuity |
| Browser and files | Signed-in browsing, local files, and generated deliverables |
Agents and Models, and Workspace Apps
Claude Code, Codex, and the built-in holaOS agent can work inside the persistent workspace. Changing the active agent does not require rebuilding the project’s skills, app connections, MCP configuration, or durable memory.
Model selection is independent from agent selection. You can use Holaboss-managed model access, your own OpenAI or Anthropic credentials, OpenRouter, or a local model.
Workspace apps extend the agent beyond chat. A HolaApp can open a URL, connect an MCP server, and expose actions to the active agent.
Shared Memory and Local Data
holaOS stores durable memory locally as readable files and keeps runtime metadata in a local state database. Workspace knowledge, preferences, identity information, and short-term session continuity use distinct storage areas.
Standing workspace instructions live in AGENTS.md, while reusable skills live under skills/. Memory access follows workspace boundaries, and persistent context can be inspected directly on the machine.
Local storage does not keep every request on the device. Calls to OpenAI, Anthropic, OpenRouter, Holaboss services, and connected cloud integrations are sent to those services.
A local model can handle inference on the machine when the workspace does not depend on hosted model access.
Install holaOS
Run the one-line installer below to install missing Git and Node.js dependencies, prepare the desktop runtime, verify the setup, and launch Electron:
curl -fsSL https://raw.githubusercontent.com/holaboss-ai/holaOS/main/scripts/install.sh | bash -s -- --launchManual Install
For manual installation, use Git, Node.js 24 or newer, and npm.
git clone https://github.com/holaboss-ai/holaOS.git holaboss-ai
cd holaboss-ai
npm run desktop:install
cp apps/desktop/.env.example apps/desktop/.env
npm run desktop:prepare-runtime:local
npm run desktop:typecheck
npm run desktop:devPros
- Persistent workspace state across agent sessions
- Claude Code, Codex, and built-in agent access
- Local files, memory, skills, apps, and MCP
- BYOK and local model options
- Free self-hosting with source access
Cons
- External provider calls leave the local machine
- Hosted features depend on Holaboss services
FAQs
Can holaOS run without a Holaboss login?
Yes. Local desktop development, local runtime packaging, and local workspace operations can run without a Holaboss login. Hosted account features and backend-connected Holaboss services require sign-in.
Does holaOS support Linux?
Linux and WSL can run the source installer for local OSS desktop development. Packaged desktop builds are available for macOS and Windows.
Can a company use holaOS internally without a commercial license?
Yes. Internal commercial use within one organization is allowed, including multiple workspaces. Third-party hosted services and commercially distributed products that embed holaOS source code require written authorization or a commercial license.
Do my own API keys consume Holaboss model credits?
No. Model calls made with your own keys consume zero Holaboss model credits. Integration calls to connected services can still use the Holaboss credit balance.
Alternatives & Related Resources
- AionUi: Free Open-source Cowork App.
- Waku: Run Claude Code, Codex & More in One Native App.
- bb: Free Open-Source Agentic IDE for AI Coding Agents.
- Odysseus: Free Self-Hosted AI Workspace with Chat, Agents, and Email.
- Agent Skills Specification: SKILL.md Format, Fields, and Directory Structure.
