holaOS wants to be the room your agents work in, not another agent
holaOS climbed GitHub trending today at 6.5k stars, and the interesting thing is what it refuses to be. It's not another agent. It's a local-first workspace that runs Claude Code, Codex, and its own native agent side by side, sharing the same memory, tools, and integrations.
Memory is the load-bearing piece. Context and project history live locally as editable files, not in somebody's cloud, and they persist across sessions and across agent switches. That last clause is the point. Right now every agent you use has its own amnesia, and moving from Claude Code to Codex means re-explaining your project from scratch. Making the memory a folder on your disk that any agent can read turns switching agents into a routine act instead of a migration.
The rest is competent and expected: 100-plus one-click OAuth integrations covering Gmail, Slack, GitHub, Linear and Notion, MCP server support for your own tools, reusable skills bundled into workflows that deploy across any agent, Opus 5 and GPT-5.6 built in or bring your own key. HolaApps is the part that isn't expected, opening real interfaces like Notion or a browser next to the agent so you can work alongside it rather than typing at it. TypeScript runtime, Electron desktop for macOS, Windows and Linux, self-hostable, with an enterprise SSO tier. License is a modified Apache 2.0 with conditions on commercial distribution, so read it before you build a business on top.
Zoom out and this is the fourth serious swing at the cockpit layer in about two weeks: T3 Code, OpenChamber, Paperclip, now holaOS. Four independent teams reaching the same conclusion, that the model is settled and the container is not, is a signal worth taking seriously. The specific bet holaOS makes is anti-lock-in, and it's aimed squarely at the fact that the strongest agents come from companies who would very much prefer to also own your workspace.
github.com/holaboss-ai/holaOS.
← Back to all articles
Memory is the load-bearing piece. Context and project history live locally as editable files, not in somebody's cloud, and they persist across sessions and across agent switches. That last clause is the point. Right now every agent you use has its own amnesia, and moving from Claude Code to Codex means re-explaining your project from scratch. Making the memory a folder on your disk that any agent can read turns switching agents into a routine act instead of a migration.
The rest is competent and expected: 100-plus one-click OAuth integrations covering Gmail, Slack, GitHub, Linear and Notion, MCP server support for your own tools, reusable skills bundled into workflows that deploy across any agent, Opus 5 and GPT-5.6 built in or bring your own key. HolaApps is the part that isn't expected, opening real interfaces like Notion or a browser next to the agent so you can work alongside it rather than typing at it. TypeScript runtime, Electron desktop for macOS, Windows and Linux, self-hostable, with an enterprise SSO tier. License is a modified Apache 2.0 with conditions on commercial distribution, so read it before you build a business on top.
Zoom out and this is the fourth serious swing at the cockpit layer in about two weeks: T3 Code, OpenChamber, Paperclip, now holaOS. Four independent teams reaching the same conclusion, that the model is settled and the container is not, is a signal worth taking seriously. The specific bet holaOS makes is anti-lock-in, and it's aimed squarely at the fact that the strongest agents come from companies who would very much prefer to also own your workspace.
github.com/holaboss-ai/holaOS.
Comments