
ASI:One
ASI:One is a personal AI from Fetch.ai that learns your preferences, remembers your conversations and talks to other people's AIs to settle plans on your behalf. It runs on the web, iOS and Android, with an OpenAI-compatible developer API.
What is ASI:One?
ASI:One is the personal AI assistant published by Fetch.ai, and its distinguishing bet is not the chat window itself but what happens between chat windows: your assistant is built to talk to other people's assistants. The terms of service call it an Artificial Social Intelligence platform. In practice you shape a single AI through ordinary conversation, it picks up your tone, your preferences and your history, and it then represents you inside a network of agents, settling a night out, a group trip or a collaboration while you are offline.
Six capabilities are advertised: a personal assistant that remembers, a social profile that stands in for you, agent-to-agent conversations, collaboration with specialist agents drawn from Fetch.ai's Agentverse and Flockx, a personal knowledge base built from your own exchanges, and two-way voice through speech-to-text and text-to-speech. Around that core sit named features: persistent memory, multi-user chat, projects, recurring and long-running tasks, rules that constrain how your AI behaves, a community skills library you can draw from or contribute to, podcast roundtables, and AI-to-AI payments backed by an integrated crypto wallet.
Connected services extend its reach: web search, Google Calendar, Gmail and Google Contacts, plus a Web3 module for balances, token swaps and transfers. Interoperability is documented rather than implied, since the publisher exposes an llms.txt summary and an agent card at /.well-known/agent.json declaring four skills, text and audio modes, streaming and bearer authentication.
There is a second audience. ASI:One is also a developer platform whose API is OpenAI-compatible on /v1/chat/completions and /v1/responses, so an existing client is repointed in one line. Three models share that surface: asi1, the general default, with a configurable reasoning budget, image input and a 196,608-token context window; asi1-ultra, the most capable, at 202,752 tokens; and asi1-mini, the fastest and lightest, at 262,144 tokens. A planner mode breaks a complex request into steps executed against agents and tools. A separate business offering deploys the same model inside an organisation, grounds it in company data with cited answers, and puts a verified brand agent in front of customers. The consumer product runs on the web and as iOS and Android applications.
What it does
- Hold a text or voice conversation with an AI that remembers your preferences across sessions
- Let your AI negotiate directly with other people's AIs to settle a plan
- Read and manage your schedule through a connected Google Calendar
- Draft emails and resolve contact names through Gmail and Google Contacts
- Run a web search when that capability is switched on
- Activate community skills or write rules that constrain how your AI behaves
- Check balances, swap tokens and send transfers through the Web3 module
When to use ASI:One / When not to
A quick filter to help you decide if ASI:One is the right fit.
When to use ASI:One
- Freelancers and founders who carry all their own scheduling and follow-up
- Socially busy people who constantly organise outings, trips and group plans
- Developers who want an OpenAI-compatible API that can call specialist agents
- Social media and community managers delegating monitoring and routine replies
- Web3 users who want wallet balances, token swaps and agent-to-agent payments in a chat
When not to use ASI:One
- Buyers who need a published price before signing up, since no tier amount is public
- Teams requiring a non-English interface, as English is the only documented language
- Organisations with EU data-residency requirements, since processing happens in the United States
- Procurement processes that demand SOC 2 or ISO 27001, neither of which is published
- Anyone under 16, whom the terms of service explicitly exclude
How to use ASI:One
A typical end-to-end flow, from setup to results.
- Create an account on the website, or install the iOS or Android application
- Go through the onboarding, which shapes your AI's personality by conversation
- Pick a voice from the catalogue under the Voice settings, then save the change
- Open Connected Services and switch on web search
- Connect Google Calendar and Contacts so your AI can handle scheduling
- Enable the Web3 module if you want wallet balances, swaps and transfers
- Write rules that set the boundaries your AI must respect
- Activate skills from the community library, or author your own
- Schedule recurring and long-running tasks, and let your AI coordinate with others
- For development, create an API key and point an OpenAI client at the ASI:One endpoint
Pros & Cons
Pros
- Training on your content is strict opt-in and off by default, with consent you can withdraw
- The opt-out control is documented and reachable from profile settings
- The API is OpenAI-compatible, so an existing client is repointed in a single line
- Three interchangeable models with context windows from 196,608 to 262,144 tokens
- A permanent free tier, plus published iOS and Android applications
- Data controller, postal address and Article 27 representative are all named in plain text
- Interoperability is documented through a public agent card and an llms.txt summary
Cons
- No price is published: the Plus and Pro tiers appear only after you sign in
- There is no pricing page at all, and no amount anywhere in the front-end code
- The site is entirely JavaScript-rendered, so nothing is readable without a browser
- English is the only documented interface language
- No subprocessor list is published
- No security certification is shown, neither SOC 2 nor ISO 27001
- No EU Article 27 representative and no EU hosting option are offered
Pricing & Plans
A permanent free plan exists: the structured data published on the homepage declares an offer at 0 USD, and the mobile application is free to download with no in-app purchase listed. Paid tiers named Plus and Pro are confirmed in the application itself and are billed through Stripe, and usage carries a configurable spend limit. No amount is published: the pricing endpoint requires authentication, and no figure appears anywhere on the public site. The lowest paid entry point therefore cannot be stated. Business and enterprise arrangements, which come with a data processing agreement, are handled through the sales contact.
- 0 USD
- permanent free tier declared in the homepage structured data
- paid tier
- amount not published
- paid tier
- amount not published
- covered by a data processing agreement
- priced on request through sales
Data, GDPR & hosting
A consolidated view of how ASI:One handles your data.
GDPR overview
The privacy policy, effective 24 June 2026, is written specifically against the UK GDPR. Concrete measures are named rather than alluded to: Fetch AI Inc is identified as data controller with its Grand Rapids address; uVue Ltd, 18 Langton Place, Bury St Edmunds, Suffolk, IP33 1NE, is the designated representative under Article 27 of the UK GDPR; lawful bases are enumerated as contract, legitimate interest, legal obligation and consent; access, rectification, erasure, portability, restriction, objection and consent withdrawal are all listed, with a one-month response deadline extendable by two months. Transfers out of the UK rely on the IDTA or Standard Contractual Clauses with the UK Addendum, and complaints may go to the ICO. No representative is designated under Article 27 of the EU GDPR, and the policy does not address EU supervisory authorities.
Who owns the data?
You keep ownership of everything you provide, including what you share with your AI. In exchange you grant Fetch AI Inc, a Delaware corporation and the named data controller, a licence to use, process and store that content in order to run and improve the service. Your AI may pass relevant information to other users' AIs once you enable that function, which is the mechanism the product is built on rather than an incidental setting. Beyond that, data may be shared with service vendors, disclosed where the law requires it, or transferred in a merger or acquisition. The publisher states plainly that it does not sell personal information to third parties.
Reuse rights
You may reuse what your AI produces without asking permission, but the terms attach real conditions. You remain responsible for what you share, for how your AI is configured, and for any action it takes in your name, including messages it sends to other users' AIs. You may not use the service to train, distil or extract the weights, parameters or capabilities of the underlying models without written permission, and you may not attempt to bypass its safety filters. When you publish AI-generated content outside the service, you are asked not to present it as entirely human-made. A detailed prohibited-content list covers sexual imagery, material involving minors, disinformation, weapons and malware.
Data retention & training
Hosting summary
Fetch AI Inc, a Delaware corporation with its headquarters at 99 Monroe Ave NW, Grand Rapids, MI 49503, is the controller. The privacy policy states that information may be transferred to and processed in countries other than your own, including the United States, and the business pages describe the top model as US-hosted. Transfers out of the United Kingdom rely on the UK International Data Transfer Agreement, or on Standard Contractual Clauses with the UK Addendum. No other hosting country, no broad region and no data-residency option is named anywhere. The domain resolves to an Amazon anycast node, which indicates content delivery rather than a storage location and should not be read as one. No security certification is published and no subprocessor list is offered, although Stripe, Google Analytics and Google reCAPTCHA are named individually as third parties in the policy text.
Things to keep in mind
Risks and trade-offs to weigh before adopting ASI:One.
- Delegating social contact to an AI can thin out the relationships it is meant to maintain
- The publisher itself warns the AI can be wrong and must not replace professional advice
- Persistent memory fed by your private life keeps widening the sensitive data it holds
- Granting Google Calendar, Gmail and Contacts access is broad, and revoking it is on you
- Crypto transactions are irreversible and the publisher accepts no liability for lost assets
- A training opt-in accepted during onboarding is easy to forget and worth re-checking
- Processing happens in the United States with no EU hosting option and no public prices
Setup & Integrations
Technical difficulty
Low for everyday use. You create an account and a conversational onboarding shapes the assistant, with no technical skill required; connecting Google Calendar and Contacts takes a few clicks under Connected Services. Voice works best in desktop Chrome or the mobile app. For developers the lift is equally small: generate an API key and repoint an existing OpenAI client, a one-line change. The one genuinely demanding part is the Web3 module, where wallets, token swaps and irreversible transfers assume real prior knowledge of crypto.
Deployment
Apps stores
Integrations
Supported languages
Behind ASI:One
Fundraising
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What is ASI:One?
Is there a free plan?
How much do the paid tiers cost?
Are my conversations used to train the models?
Is there an API?
Which platforms is it available on?
Is there a minimum age?
Who is legally responsible for my data?
Which law governs the service?
What can it connect to?
Should you pick ASI:One?
ASI:One is a young product resting on an established base. The domain was registered in February 2025 and first archived days later, yet it belongs to the Fetch.ai ecosystem alongside Agentverse and Flockx, and it inherits that ecosystem's agent infrastructure. The idea it stakes itself on is genuinely distinct from the assistant crowd: rather than answering you faster, it negotiates with other people's assistants on your behalf, which is either the feature that sells it or the reason you will keep it switched off.
On governance the record is better than average. Training on your conversations is strict opt-in, off unless you turn it on, adjustable in profile settings, and the consent can be withdrawn. The data controller, its postal address and an Article 27 representative are named in plain text. Retention periods are itemised rather than gestured at.
Two reservations deserve weight. The first is pricing: paid Plus and Pro tiers demonstrably exist, billed through Stripe, but no amount is published anywhere a prospect can see it before creating an account. Evaluating cost of ownership up front is simply not possible. The second is jurisdictional reach: the privacy policy is written for the UK GDPR, the named representative is a UK one, processing happens in the United States, and no EU representative or EU hosting option is offered. An EU buyer with data-residency requirements will not find them here.
Add to that a single documented interface language, no published subprocessor list and no security certification, and the profile becomes clear: an inventive consumer product with a credible developer platform attached, best suited to individuals and small teams willing to trade procurement guarantees for a genuinely new way of delegating coordination.
- Choosing a selection results in a full page refresh.
- Opens in a new window.