Watch the actual Halo desktop app read a project file and prepare a shift handoff. The device activity feed records the file read alongside the conversation.
Read a project notes file and prepare a three-bullet shift handoff.
Halo read the file through its desktop tool and returned a handoff. The activity feed showed read_file with an OK result.
The app remained disarmed. This demonstration used a read-only task and a fictional project file.
The demo file records 24 connected and labeled workstations, Wi-Fi checks in three meeting rooms and an unconfirmed access-point delivery. Halo reads that file, then summarizes completed work, the delivery follow-up and the fact that no customer communication has been sent.
Captured September 7, 2026 · Real application output, fictional project data. Hosting, models and connected tools depend on the deployment.
A useful answer is often only the beginning. The next step is in a project folder, an application or a system someone has to check. Halo is our approach to connecting the conversation to that work, with a desktop client that gives the assistant a controlled way to interact with a device.
Halo connects outward from the desktop to an OpenClaw gateway. The gateway runs the agent conversation and requests available tools; the desktop client performs supported device actions and returns the results. This separates the interface on the machine from the service coordinating the task.
The chat sits beside a device activity feed. In the recorded build, Halo reads a specific file, receives its contents and writes the handoff in the conversation. The successful file read appears in the activity feed, so there is something concrete to inspect alongside the answer.
The desktop build includes an arm control and a stop mechanism. Read-only context can be available while disarmed; state-changing device actions require arming. The demonstration stays disarmed. A customer rollout still needs deliberately scoped tools, accounts and access.
Start with a repeatable task such as reading selected project notes and preparing a shift handoff. Agree on the input folder, the expected output and the person who reviews it. Expand to more connected work only after the first workflow behaves reliably.
The goal is to make the path from a request to a useful result shorter, while keeping the work understandable to the person responsible for it.
These are development goals, not a list of released features or promised delivery dates. Customer needs and testing will determine what we build and release next.
Make connecting a new device, understanding its access and recovering a lost connection clearer for customers.
Improve progress reporting, interrupted-task recovery and continuation between connected surfaces, with clear confirmation of what actually finished.
Develop tested connections and reusable procedures for common operational tasks, with explicit limits on what each workflow can reach and change.
Halo is the GHT agent layer, an AI operator that doesn’t just chat, it acts. Through a secure gateway it can work across your desktop, your phone, your servers, and your tools, carrying one conversation everywhere you are.
Ask it to check a server’s disk and restart a service if it’s healthy, and it runs the steps, shows its work, and reports back. Every action is routed through a gateway you control, and Halo is built on the OpenClaw runtime.
Actual captures from Halo, using fictional project information. Click a screen to inspect the application at full size.
Halo runs real steps across your tools and devices, shows each tool call, and reports the result.
Desktop and Android share one operator, start a task on your laptop, follow it up from your phone.
Every action routes through a gateway you control, so you decide what Halo can reach.
Connects to the OpenClaw gateway. Available models, tools and access are configured for the deployment.
Tell us what you’re trying to build and we’ll get you set up, or point you to the right download.
