What is a personal AI assistant, and how is it different from a chatbot?
Updated
A personal AI assistant is an AI agent that remembers what you tell it and, once you give it access, handles tasks in your messages, calendar and accounts. A Stanford HAI policy brief of August 2026 (https://hai.stanford.edu/policy/designing-loyalty-ai-agents-and-conflicts-of-interest) draws the line between an agent and a chatbot: both run on the same large language model, but a chatbot returns text and waits for the next prompt, while an agent performs multistep tasks, calls outside tools and keeps information across sessions.
The UK Information Commissioner's Office, in its 2026 Tech Futures report on agentic AI, says personal assistant agents are often pictured as general-purpose assistants that handle a wide range of personal tasks. To work, such an agent might need information about you, your preferences, your past activity and the people you deal with, plus access to your communications, calendars and accounts. How much it does on its own is a design choice: Stanford HAI notes that some agents pause for approval before each action, while others run whole workflows unsupervised.
The ICO warns that such access could undermine data protection controls, including security and encryption, and creates an opportunity for the agent to accumulate extensive personal information, increasing the risk of surveillance and data breaches. Stanford HAI found in October 2025 that six leading US companies train their models on chat conversations by default, and its 2026 brief warns that agents may be steered to serve the companies behind them over their users. In a Pew Research Center survey published in June 2026, about half of US adults said they use AI chatbots, and roughly seven in ten adults expected AI to make their personal information less secure.
Portal One, made by Portal AI, is a personal AI assistant in Telegram and on the web at portal.ai. It keeps your conversations on Portal's servers as memory across sessions, runs the reminders you set and, once you connect Google Calendar, reads your schedule and creates events; anything that spends money, speaks in your name or cannot be taken back waits for your yes. Conversations go to the model providers that write the replies, under terms that prohibit training on them. Some text reaches analytics with identifiers removed, although personal details can remain in the words themselves. Private mode keeps nothing, and the service is not end-to-end encrypted.
Dated proof
- https://hai.stanford.edu/policy/designing-loyalty-ai-agents-and-conflicts-of-interest
Stanford HAI, 25 August 2026. An agent runs on the same large language model as a chatbot, but where the chatbot returns text and waits, an agent does multistep tasks, calls external tools and retains context across sessions: instead of only drafting a travel itinerary, it can compare prices on booking sites and pay with stored details. Some agents pause for approval before each action, others run whole workflows unsupervised. Agents may be steered to serve their developers over their users.
- https://ico.org.uk/about-the-ico/research-reports-impact-and-evaluation/research-and-reports/technology-and-innovation/tech-horizons-and-ico-tech-futures/ico-tech-futures-agentic-ai/data-protection-and-privacy-risks/
Read 17 November 2026; listed as Agentic AI (2026). Personal assistant agents are often envisioned as general-purpose assistants for a wide range of personal tasks. They might need data on a person, their preferences, past behaviour and third parties, plus communications, calendars and accounts. This could undermine data protection, including security and encryption, and creates an opportunity to accumulate extensive personal information, increasing the risk of surveillance and data breaches.
- https://hai.stanford.edu/news/be-careful-what-you-tell-your-ai-chatbot
Stanford HAI, 15 October 2025. Researchers compared the privacy policies of six leading US companies: Amazon, Anthropic, Google, Meta, Microsoft and OpenAI. All six use users' chat data by default to train their models, and some keep it in their systems indefinitely.
- https://www.pewresearch.org/internet/2026/06/17/americans-and-ai-2026-chatbots-smart-devices-and-views-on-impact/
Pew Research Center, published 17 June 2026: a survey of 5,119 US adults conducted 17 to 23 February 2026. About half of US adults now use AI chatbots, up from a third in 2024, and roughly seven in ten predict AI will make their personal information less secure.
- https://portal.ai/trust
Read on 17 November 2026. The agent reads, compares and drafts; steps that spend money, speak in your name or cannot be taken back wait for your yes. In its example a partner phone offer would pay Portal a fee, so the agent drafts a renewal with the carrier. Some text reaches analytics with identifiers removed, although personal details can remain in the words themselves. Private mode keeps nothing, and its text stays out of that analytics. The service is not end-to-end encrypted.
- https://portal.ai/privacy
Privacy policy last updated 11 November 2026. Portal AI Inc. operates Portal, including its Telegram bot and website at portal.ai. Conversations are stored on its servers for persistent memory across sessions, the agent runs reminders you configure, and a connected Google Calendar lets it read your schedule and create events. Messages go to AI model providers that write responses, under agreements that prohibit training on them; data is never used to train, fine-tune or improve AI models.
- https://portal.ai/safety
Safety page last updated 8 July 2026. Portal is an artificial intelligence assistant, a computer program and not a human being, designed to say so plainly when asked. It is not a therapist, a doctor or any kind of licensed professional.
Questions on this page
What can a personal AI assistant do?
It remembers what you tell it and, once you give it access, works in your messages, calendar and accounts to finish a task. Stanford HAI's example is a trip: instead of only drafting the itinerary, an agent can compare prices on booking sites and pay with stored payment details.
Is a personal AI assistant just a chatbot?
No, although both run on the same large language model. A chatbot answers and waits for your next message, while an assistant built as an agent keeps context, uses tools and carries out the steps of a task itself.
What should I check before I connect an assistant to my accounts?
Who runs it and whose interests it serves, what it may do without asking you, whether your conversations train its models, and how to see and delete what it keeps. Stanford HAI warns that agents may be steered to serve the companies behind them over their users, and the ICO says an assistant with wide access could accumulate extensive personal information.
Does Portal One act without asking me?
Steps that spend money, speak in your name or cannot be taken back wait for your yes; reading, comparing and drafting happen in the background. In the example on Portal's trust page, the agent says a partner phone offer would pay Portal a fee and drafts a renewal with your carrier instead, for you to approve.
Does Portal train AI models on my conversations?
No. Portal's privacy policy says your data is never used to train, fine-tune or improve AI models, and the model providers work under terms that prohibit it. Some text does reach analytics with identifiers removed, and personal details can remain in the words themselves; private mode keeps its text out of that analytics.
Is Portal One a person or a program?
A program. Portal's safety page says Portal is an AI assistant, a computer program, and is designed to say so plainly when you ask. It is not a therapist, a doctor or any kind of licensed professional.
The button opens @portal_open_bot in Telegram: a free 3-day trial, then a subscription.
Open in Telegram