holaOS is a desktop workspace, built as an Electron application for macOS, Windows, and Linux, in which an agent works beside the applications and tools a team already runs. Apps installed from an in-workspace marketplace open as live, interactive surfaces next to the agent, so you watch it act inside an app and can take over at any point; what you click, type, or open in that interface becomes context the agent already has. Any URL plus an MCP server can be pointed at to create a new app.
Several agents run in the same workspace over the same tools, files, and memory: Claude Code, Codex, and the built-in holaOS agent. Memory is a single shared store kept locally as plain files you can read and edit, structured and embedded so context returns across sessions and across whichever agent is driving.
Capabilities are added rather than built. One-click OAuth integrations connect Gmail, Notion, Slack, GitHub, Linear and more; any Model Context Protocol server can be plugged in; skills package a workflow once for any agent to run; and combos bundle skills and integrations into a single install. Chat tools including Slack, Feishu, DingTalk, and WeChat are connected so the agent reads decisions where they were made, with access granted per tool and per scope. Agents also drive a signed-in browser, generate images, video, and audio, write .xlsx, .pptx, and .docx files, and run on a schedule or a trigger.
Models are either built into the product through its own gateway — Kimi K3 and GLM 5.2 for everyday volume, GPT 5.6, Claude Opus 5, and Fable 5 for harder problems — or supplied as your own keys for OpenAI, Anthropic, or any OpenAI- or Anthropic-compatible endpoint, billed to your own account. Workspace files, memory, embeddings, and session history stay on local disk, and each action the agent takes is approved by the user. HolaHub publishes runnable agent recipes — model, prompt, skills, MCP servers, integrations — that install into another workspace in one step. An enterprise path adds single sign-on with per-role permissions, audit logs, and on-premise or private-cloud deployment.
Features
- HolaApps: marketplace apps open as live surfaces beside the agent, with context flowing both ways
- Multiple agents: Claude Code, Codex, and the built-in agent share one workspace, tool set, and memory
- Shared local memory: plain readable files, structured and embedded for recall across sessions and agents
- Integrations: one-click OAuth to Gmail, Notion, Slack, GitHub, Linear, and more
- MCP: community or self-hosted MCP servers add tools to every agent in the workspace
- Skills and combos: packaged workflows, and bundles of skills plus integrations installed in one step
- Chat tools: Slack, Feishu, DingTalk, and WeChat connected as context sources, approved per scope
- Agent-driven browser: signed-in browsers agents navigate, click through, and extract from
- Generation and deliverables: image, video, and audio models, plus .xlsx, .pptx, and .docx output files
- Automation: scheduled and triggered runs that produce digests, monitors, and reports on their own
- Model choice: built-in frontier models through the gateway, or your own OpenAI- or Anthropic-compatible keys
- HolaHub: publish and install agent recipes covering model, prompt, skills, MCP servers, and integrations
