Open Interpreter’s 01 Lite: A Voice Interface for Computer Control

The team behind Open Interpreter, an open-source project for controlling a computer with natural language, has introduced a hardware device called the 01 Lite. Matthew Berman sees it as an open-source alternative in the emerging category of devices such as Rabbit and the AI Pin. He first plays the project’s launch film, then demonstrates the underlying software himself.

In the promotional film, a project presenter argues that computers should be as easy to interact with as people. She describes an open, flexible foundation for the next generation of computing, with a language model at the center of the operating system. This is the team’s vision for its product, rather than an independent evaluation of it.

The presenter describes the 01 Lite as a device that listens when a user presses a button and operates a computer to carry out requests. She says the language model can run on a home computer or on the project’s cloud server, with Wi-Fi or a hotspot making the device portable. Local and hosted operation are presented as options, not as proof that every deployment works offline.

In the film’s demonstration, she asks for the next sunny day and then whether she has anything scheduled. The assistant checks the calendar on the connected computer. She asks about concerts in Seattle, chooses a Bad Bunny show, and asks it to add the event with a ticket link and text the link to Page; the assistant reports that both actions are done.

The assistant then mentions an email from Mike about an article. The presenter says the PDF is on her desktop, and the assistant finds it and asks whether to send it. She notes that she could also ask the device to read or edit the document, but instead confirms the email; the assistant reports that it has sent the file.

Back at the laptop that the device has been controlling, the presenter says it cannot do everything out of the box. To teach a new workflow, she guides it through opening Slack, using Command-K, selecting a recipient, clicking the message box and sending a sample message. At each step the assistant asks whether it performed the action correctly, and she confirms before it saves the skill.

The presenter suggests that similar instruction could teach other workflows across applications and websites. She then asks the assistant to monitor her email for invoices, inspect their attachments and send relevant details to Mike on Slack. The launch film shows a test invoice being sent and the assistant reporting that it messaged Mike. This is a staged product demonstration, so it does not establish reliability for arbitrary workflows or unattended monitoring.

At the 2024 launch, the team advertised a $99 preorder and said the version customers received would work as a standalone device using its hosted service. It announced developer previews of the 01 Lite, 01 server and 01 OS, along with software, CAD files and wiring diagrams for builders. Prices, availability and capabilities here are historical statements from the announcement.

The presenter imagines a wider ecosystem of open AI devices, including toys and robots, and says the team intends to release a computer-control language model, phone app and fully offline handheld device in the following weeks. Those are plans stated in the launch film, not a verification that the products shipped on that schedule or worked as envisioned.

Berman distinguishes Open Interpreter, the underlying project for natural-language computer control, from the 01 project’s voice interface. He says he spent several hours trying to get 01 OS working but could not, possibly because of his computer. His following demonstration is therefore of Open Interpreter itself, not a successful hands-on test of the 01 Lite or 01 OS.

For his 2024 setup, Berman uses Conda to create a Python 3.9 environment named oi, activates it, installs the package with pip install open-interpreter and starts it with interpreter. He says the program requests a GPT-4 API key and that other language models can also be used. Installation steps, model support and service requirements may have changed since the recording.

Berman first asks which folders are on his desktop. Open Interpreter shows the command before running it, and he approves it; the terminal then lists the folders. He next asks it to open a new Chrome tab, reviews and approves the generated action, and shows the tab opening. The approval step is meaningful because generated code acts with the user’s computer permissions.

A more ambitious request asks it to open Gmail and find new email from the preceding 30 minutes. It opens Gmail and generates a script, but the script encounters an error after Berman approves it. He then asks for the latest post in his Twitter timeline: the software opens Twitter, but says it cannot directly access the timeline content because of privacy concerns. Neither request is shown completing as asked.

Berman concludes that there is still a lot of debugging to do, though he finds the software impressive. He suggests trying Open Interpreter with a local model, says he will keep working on the voice setup, and promises another tutorial if he gets it running. He also describes the device as an early developer preview and quotes $109 including shipping at the time.

Looking ahead, Berman imagines people speaking directly to devices that control an operating system locally or elsewhere, and eventually using natural language and a language model for direct execution. That is his prediction, not a demonstrated capability. He says he wants to try the 01 device and compare the emerging category with his Rabbit R1.