Openai reveals “Agent Chatgpt” that gives Chatgpt its computer to use your email and web applications independently, download and create files

Openai does not leave Delay the open source artificial intelligence model Slowly charge other features.

today , The company reveals Chatgpt agentIt is a feature that allows AI Chatbot to browse the web independently, conduct comprehensive research, download and create new files for its human users using its virtual computer.

Come again? Chatgpt is now king PC? It can use this computer to log in to your user, or human user accounts, download or send things?

This is true, at least in the default sense, according to Openai. The company also shows:

“The model can choose to open a page using a text browser or a visible browser, download a web file, and process it by running an order at the station, then display the output again in the optical browser. The model adapts its approach to carrying out the tasks of speed, accuracy, and efficiency.”

How to use Chatgpt agent

Users can involve the agent by clicking on the “tools” button in the chat of ChatgPT entry, opening the drop -down menu, and determining the “agent mode” from the available options.

After that, when it is run, it describes a task in an ordinary language, and the agent can implement it through web environments and local applications, as it combines thinking with procedures that the human user had not previously been able to perform on his own device manually.

Chatgpt Agent can contact applications like your personal or commercial Gmail and GitHub, so that it can withdraw useful information – emails or software instructions – from your accounts to help with the tasks you require do. He can contact the third -party programming interfaces (API) to withdraw information and use applications and services connected to them as well.

If the web site needs to log in, you can do this safely with a special browser view, which allows the agent to dig deeper and deal with the most customized tasks, such as checking your incoming models or filling your behavior on your behalf.

Go to where the operator was unable – without contact

The new Chatgpt agent builds and expands from Opeenai agent has been released In January 2025, which allowed ChatGPT to browse the web, fill the models, apply requests, and carry out other web -based tasks in the private “cut browser”, that is, a custom web browser based on the cloud that was preserved and displayed for each operating session.

However, the operator was limited to interacting with web sites and web-based applications-and not programs that could work locally on a computer, such as spreadsheet software and deck width width.

The new Chatgpt agent can browse websites, interact with online models, operate the code, analyze data, and provide final outputs – such as editing presentations or data schedules – based on the entire user instructions.

The disclosure of a report published a few days ago comes through the independent subscription technology industry site Information This indicates that Openai will upgrade Chatgpt To be a direct competitor to the Microsoft desktop applications (for example, Excel, Word, PowerPoint, etc.)

Merging the operator and deep research into one agent

In fact, Openai places a Chatgpt client as a merger of two former agents – operator and deep research, The latter was presented in February 2025Which is comprehensively looking for the web through its uninterrupted browser only to find information and assemble it in long and in -depth reports (and thus name). Openai also writes in the blog post:

“The operator was unable to dive into the analysis or write detailed reports, and deep research was unable to interact with web sites to improve results or access content that requires user authentication. In fact, we saw that many information users who tried with the operator were more suitable for deep search, so we collected the best in each of them.”

The gradual elimination of the previous operator tool will be eliminated, but users can still access deep -drop research at the ChatGPT interface.

Whether the use of a visible browser to interact with a website or station to run the Python icon, the agent moves smoothly between the tools inside one session.

It supports a wide range of use cases, from competitors analysis and reports to planning trips, summarize emails, or reservation dates.

Users can boycott, stew or stop the task at any time, picked up the agent directly as it stops.

Availability and access

Starting today, subscribers of that of $ 200 will have a “Pro” full access to a Chatgpt agent, with a monthly share of 400 messages.

Chatgpt Plus ($ 20 per month) and the team ($ 30 per month) will be able to reach during the next few days, with 40 messages per month. Additional use is available through credit -based options.

Openai said in a joint version with Venturebeat under the siege that Chatgpt Enterprise subscribers and education will be able to access the feature in the coming weeks.

At the present time, this feature was not yet available in Europe or Switzerland, and the population who disappoints there is no doubt.

It was built with safety and control in the foreground

Given that the agent can now take action on behalf of users-including on registered websites or connected applications-enter OpenAi large-scale safety measures.

These include user assurances before taking action, active supervision of sensitive tasks, and technical guarantees to reduce unintended behavior.

The main protection includes:

• Claims of confirmation Before procedures such as sending models or sending emails
• Put the watchWho stops executing when the user becomes inactive
• Refuse highly dangerous tasksLike financial transfers or privacy violations
• Not to keep memory During the agent sessions

Low domain classification

Looking up with its preparation framework, Openai is treated Chatgpt agent as a file High capacity The system is in the biological and chemical fields.

Although there is no direct evidence of misuse, the company stimulates the strongest safety guarantees with caution.

This includes improved rejection, red groups by biomedice experts and improved detection systems.

Remember that manRecent information has been released about the new Claud Opus 4, and Other polls of advanced artificial intelligence modelsShe explained that when accessing access to tools and external applications such as e -mail, they can in some cases take action that they believe are ethical and ethical, but this may expose the user to settle for the user, such as sending government agencies or journalists who make a mistake in part of the user.

The model believes that it behaves like “infrinists”, but in reality, it may offer the privacy of the user, security and private information and alerting authorities to the absence of anything, or where it is doubtful.

Strong performance in the tasks in the real world

Agent Chatgpt was not more theoretical-he has achieved strong results through a number of criteria designed to simulate knowledge works in the real world. It has set a new high degree on the last humanity exam with 44.4 using parallel deportation methods, and achieved 27.4 % on the difficult Frontiermath standard.

In the spreadsheet, record 45.5 % – more than the COPILOT in Excel Performance.

Current restrictions and the following steps

Some features, such as the generation of slides, are still in beta and you may feel essential in coordination or slightly vary between examining within the application and exported files.

Openai trains the next repetition of this feature to improve Polish and planning.

Agent Chatgpt’s launch indicates a shift in how users interact with artificial intelligence – from asking questions to set full tasks.

Thanks to its ability to think, act, and produce products, Openai is betting that users are increasingly wanting Amnesty International not only to help them, but work for them. While the company emphasizes that the agent is still developing, he believes that this launch is the basis for a more interactive future directed towards artificial intelligence.

Daily visions about business use cases with VB daily

If you want to persuade your boss at work, you have covered VB Daily. We give you the internal journalistic precedence over what companies do with obstetric artificial intelligence, from organizational transformations to practical publishing operations, so that you can share visions of the maximum return on investment.

Read with us privacy policy

Thanks for subscribing. Check more VB news bulletins here.

An error occurred.