An AI agent and its subagents work in your editor, terminals and browser, next to you. Use any model, or run one on your own computer. Claude Code plugins, skills and agents run as they are.
Sign in or paste an API key, then pick the model for each chat and for background work like titles and recaps. A local model keeps what you send on this computer. With OpenRouter connected, quick choices such as the browser’s next click go to a small, fast decision model instead of the chat’s model, with a fallback when it has no answer.
The agent works inside your project folder, with its files, terminals and Git, and you work next to it in the same window. It can open the page it builds in the Browser, beside you. Link Claude Code, Codex, Gemini CLI or another installed coding agent to a chat, and its terminal opens beside the conversation.
Files open in a full code editor, with completion, hover, parameter hints, errors underlined as you type, go to definition and rename across files. Language servers install themselves, for the languages Localix supports, the first time you open one of their files. Predicting your next edit is optional: it is off by default and uses model usage.
The agent hands a self-contained task to a subagent, a session of its own that works while the agent goes on and answers back when it is done. Each one shows what it was asked and what it answered, and sits under its chat in the sidebar, where you can open it and talk to it directly.
Find, install and update mini apps, plugins, skills, agents and MCP servers in one tab, from the Localix marketplace, Localix Community, Anthropic’s marketplaces and any you add under Sources. Search matches by meaning and reaches inside packages. A plugin lists what it adds and what runs on your computer before you say yes. Most of what you wrote for Claude Code loads as it is, with no conversion.
A mini app is an application inside Localix: a screen, tools the agent can call and a guide for using them. Localix’s own mini apps are in the Localix marketplace and install from the Store like any other. Describe one to the agent, with the authoring plugin from the marketplace, and it writes, checks and installs it; through the host it gets decision models, embeddings and search. On macOS it also installs as a Mac app, in the Dock, Cmd+Tab and Spotlight.
83 pages
v0.1.0-beta.27
Claude Code plugins now run on one shared engine, the open llx-plugin kit, so a plugin behaves in Localix as it does in Claude Code.
2 added · 7 changed · 5 fixed
v0.1.0-beta.26
Mini apps get smarter in the chat: the agent now finds things in them by meaning, and a mini app you made shows its README and screenshots in the Store.
5 fixed
v0.1.0-beta.25
A small fix release: mini app items keep their data when one field changes, the mini app authoring plugin installs again after an upgrade from an old beta, and the localix terminal command no longer asks you to fix a PATH that is already right.
4 fixed
v0.1.0-beta.24
The agent is more careful with steps that change things: a failed step stops the rest of its plan, and an action that may already have gone through is not repeated without asking.
2 added · 7 changed · 2 fixed
v0.1.0-beta.23
The agent island is back for questions and approvals, and the agent works real web pages far more reliably: dropdowns, frames, late content, long pages and booking-style forms.
5 added · 3 changed · 6 fixed · 1 removed
Free · MIT license
The app for Macs with Apple silicon. On other systems, build it from source.
In the model picker under the chat, sign in or paste an API key, or pick a model that runs on your computer.
Add a project folder and say what you want done. The agent works in it, next to you.