Your AI Got Faster. Your Team Didn't. Here's a Workstation That Tries to Fix That.
You've probably noticed the pattern by now: every AI tool you adopt makes you faster, but the hand-offs between you and everyone else still run through people. You copy one AI's conclusion into another AI, send files and findings to a colleague, and they explain it all over again to their own AI. Individuals got quicker; the team stayed the same speed. Halo is an open-source desktop AI workstation that tries to change that by giving AIs a way to divide work, hand it off, and stay on duty themselves.
What It Does
Halo is a cross-platform desktop AI workstation (macOS, Windows, Linux, plus a web version) that covers the everyday stuff you'd expect: chat, writing code, making documents, research. The chat sits next to a canvas where reports, code, web pages, terminals, and Word/Excel/PDF files open right beside the conversation. You can select a passage, add a note, and the AI knows exactly what you're referring to.
The more interesting layer is what the project calls Digital Humans — AI members that stay on duty — which you can organize into teams that split the work. A lead coordinates the team while teammates' Digital Humans work on their own computers, exchanging messages, tasks, and results through the team. When you leave your desk, you can hand off work and follow up from your phone, Feishu, WeChat, or WeCom.
Under the hood, the workstation offers multiple coding engines (Claude Code, Codex, Halo SDK and others), an AI-driven browser and terminal you can watch and take over at any moment, and built-in document skills for creating and editing Office formats without installing a pile of dependencies first.
Why It's Cool
-
It targets the real bottleneck. The README makes a sharp observation: personal productivity adds up, but collaboration multiplies. Halo is built around the hand-off problem rather than just making a single chat window smarter.
-
Memory that doesn't reset. Files, folders, PDFs, and Office documents in a knowledge base get cited with sources while you chat, and conversations in the same space share memory organized into current state, history, and topics. A new chat doesn't start from scratch.
-
You can see what the AI did. The changes view shows exactly what the AI modified, and you can kick off a quick review or a multi-role team review with one click. That's a sensible answer to the trust problem of letting an agent loose on your code.
-
Local files stay local. You can attach local files and folders directly, nothing is uploaded — a detail that matters if you work with anything sensitive.
-
The capabilities are opt-in, not buried. A "+" in the input box turns on web control, team collaboration, the AI terminal, OCR, and operating Halo itself. It's a small design decision, but it keeps the default surface uncluttered.
-
A credible dogfooding claim. The project notes it's written by Halo itself — over 500,000 lines of code, running stably in enterprise financial environments. You can take that with whatever salt you like, but it's a concrete data point rather than a vague boast.
How to Try It
- Head to the releases page and download the build for your platform — macOS, Windows, or Linux. There's also a web version if you'd rather not install anything.
- Check the website and docs for setup guidance and details on connecting external channels like Feishu, WeChat, and WeCom.
- Start simple: open a chat, attach a local folder or file, and try the canvas. Once that feels natural, turn on team collaboration from the "+" menu and see how the Digital Human setup fits your workflow.
- If you want to contribute, the repo is at github.com/openkursar/hello-halo and it's MIT licensed.
Final Thoughts
Halo is aimed at heavy AI users, developers, researchers, operators, and small teams bringing AI into daily work — and that framing is honest. If you're a solo user who just wants a chat window, you probably don't need the team layer. But if your work constantly flows between people and their AI assistants, the idea of giving those assistants a shared place to hand off tasks and keep memory is worth a look. The README is upfront that this is a large, self-improving workstation, so expect a learning curve. Still, the underlying premise — that the bottleneck moved from individual output to coordination — is one a lot of teams will recognize.
Follow @githubprojects for more developer tools and open source projects.