ChatGPT becomes a task assistant with local files and models for personal and business use

OpenAI is internally testing an update for ChatGPT on the web version that, according to the findings published in networks, will start to be deployed in the coming weeks. What appears in the code and in the internal references shows several movements that go beyond simple aesthetic improvements: the platform seems to be prepared to incorporate functions oriented to both the daily user and more complex business deployments.

Among the new features detected is a key name function Salute which in the internal signs appears as a system to create tasks that can include file uploading and allow the tracking of their progress. This opens the door to a ChatGPT that not only responds in real time, but acts as a light management tool: imagine sending a document and that the IA itself organizes subtasks, follow up and return progress is a step towards more integrated assistants in personal and professional workflows. The researcher Tibor Blaho shared catches and references of these detections in X, which put the matter in the public hearing ( see publication of Tibor Blaho).

Another interesting change in the code is a flag called "is model preferred". In the context that has been seen, this label could allow ChatGPT to automatically select an optimized model for local results - for example, business, restaurants or hotels - when maps or local information sheets are shown. That is, the experience could be made more contextual: instead of asking the user to specify, the system would choose a specialized model according to the type of query and the interface where the query appears. Such adaptation would increase the relevance of responses for local searches and commercial uses.

In the field of infrastructure, there are references to a new safe tunnel for MCP (Model Context Protocol) servers. The description that appears suggests a connection to OpenAI through a tunnel client hosted by the client, using only outgoing traffic by HTTPS, so that it would not be necessary to open incoming ports in the firewalls. For organizations that integrate models with internal data and own tools, a solution thus reduces operational friction and risks associated with direct exposure of internal services. This kind of improvement often responds to the demands of security and operations teams that call for simple and less invasive mechanisms to connect private infrastructure with cloud services.

In parallel, the company is also experimenting with online editable blocks for code and mathematical formulas, an evolution of the enriched text areas that have already been seen recently. OpenAI recently added what they describe as a Formatting block: a small editing bar that appears when selecting text within more modern drafting areas (such as mail or long text erasers). The extension of that idea to code blocks and mathematicians would mean that users can edit complex fragments without leaving the conversation flow, with better visual representation and specific editing tools; this could be especially useful for developers, students and technical professionals who use ChatGPT to review or generate code and formulas. Specialized media have already documented the first appearances of this enriched interface ( see technical coverage).

If all these pieces are combined, the picture is that of a ChatGPT that advances on three fronts: more practical utility for tasks (with Salute), automatic adaptation capabilities to specific contexts (with the preferred model flag) and security and integration improvements for companies (with the safe tunnel for MCP). All this, in addition, is accompanied by a more powerful interface to handle technical and structured text within the conversation itself.

What practical implications does this have for users? For the general public, the arrival of editable blocks and a more powerful formatting bar will make the writing, correction and editing within ChatGPT feel closer to a traditional editor, reducing steps and friction. For professionals, the possibility of uploading files to tasks and following their progress transforms the IA into an assistant who can orchestrate work rather than just respond. And for organizations, improvements in secure connectivity and the ability to target specialized models promise more robust integration and more relevant responses in business contexts.

However, these developments also bring questions that should be taken into account. At the level of privacy and governance, allowing to upload files to tasks involves decisions on storage, retention and use of data. In corporate areas, although an emerging tunnel reduces the need for network changes, it is still necessary to audit what data are shared and how the interactions between internal systems and the external platform are recorded. From a technical perspective, the automatic selection of models by context requires the company to document clearly what optimisations apply and how the bias or limitations of each specialized model are monitored.

OpenAI has been publishing updates and guides on its work and public documentation; following its official channels is the best way to know when these tests reach a general deployment and what controls they offer for companies and users. Meanwhile, the technical community and industry observers continue to look at signals in the code and in research publications to understand the platform's direction ( OpenAI - official blog).

In short, what appears in the internal tests points to a ChatGPT more oriented to real workflows and a mixed use between consumer and company: assistants who manage tasks with files, models that automatically fit the local context and connection channels designed for corporate environments. It remains to be seen when and how these capabilities will be released, and what controls and guarantees will accompany each function, but the trend is clear: interaction with the large models continues to evolve towards richer, central experiences in productive day-to-day.