FAQ

WAM Agents is a personal AI agent in Telegram with a name and avatar you choose. You tune it through an ordinary chat and add it to groups: your team, friends and clients talk to it like any other participant. It works in DMs, groups and forum topics — no separate app.

By default the agent runs on DeepSeek — included in every plan, no OAuth or third-party keys required. On Pro you can attach your own model: Claude with /claude, or Codex on your ChatGPT subscription with /codex. An optional upgrade on top of DeepSeek.

Out of the box on DeepSeek the agent handles text and code, runs bash, calls MCP tools, holds multi-turn dialogues, understands images and searches the web. All of it is included in every plan, with no extra keys.

Yes, natively. Voice is transcribed automatically (Deepgram Nova-3), the agent sees photos and screenshots, and it reads PDFs itself, going through a large document in parts. Files come in and go out right inside Telegram — no manual unzip.

Yes. The agent works in DMs, groups and forum topics. In groups the /mention on|off command controls behaviour: reply only to @mentions and replies — or to every message from the owner.

An optional upgrade — a dedicated group workspace where each topic becomes an independent session with its own memory. Handy for running several projects in parallel without mixing contexts. The agent walks you through setup step by step; the step can be skipped and picked up later.

Yes. The agent remembers your preferences, contacts and projects across chats. Memory lives inside your container and can be edited by conversation — just ask it to add, update or forget a fact. Nothing is shared with other people.

DeepSeek is the default model: no key of yours is needed, and its work is paid for with the magnets of your plan; it handles text, code, bash and MCP tools well; vision and web search run through OpenAI and Perplexity. Your own model (Claude, Codex) is an optional upgrade for the deepest reasoning and for long documents: the limit there comes from the context window of the model, not from a page count.

On Pro — two commands: /claude for Claude and /codex for Codex. Send /claude in the chat with your agent. The bot replies with an OAuth link — sign in with the provider and send the code back to the chat. After that switch models with /model. Refresh the token later with /login. Codex has a command of its own — /codex: it authorises by device code right in the chat, and /model then lists the models of your ChatGPT plan.

No. On the default plan the agent already does all the essentials; vision and web search run through OpenAI and Perplexity and are paid in magnets. Your own model is optional — for when you need a larger context window or deeper reasoning.

Free — $0: 200 Ⓜ your first month, then 100 Ⓜ every month, no card. Pro — $13.90 per month: it includes the infrastructure and 400 Ⓜ a month for everyday agent work — enough for 20 image generations, or 30 agentic tasks with web search, or 266 web searches, or hundreds of text messages — plus web search, vision, voice transcription, document handling, MCP tools, long-term memory and auto-updates. Nothing else to pay to get started.

Open @wamagentsbot in Telegram and hit Start — the bot walks you through onboarding. The Free plan needs no card and gives you 200 Ⓜ your first month, then 100 Ⓜ every month. Pro is $1.39 for your first month when you subscribe within 7 days of signing up, then $13.90/month; cancel anytime with /cancel.

The agent tells you and offers options: wait for the quota to refresh at the start of the next subscription period, top up with a magnet pack — 500 Ⓜ for $9.90, 1,500 Ⓜ for $27.90 or 5,000 Ⓜ for $89.90 (Pro only), or on Pro switch to your own model with /claude. The quota refreshes every period.

Yes. Share your invite link: after your invitee's first payment you get 200 Ⓜ that never expire — for every paying invitee, with no cap.

Send /cancel in @wamagentsbot — access stays until the end of the paid period. After that the agent moves to Free and keeps working — files, memory and purchased magnets stay where they are.

The Service is provided and licensed by Magnet Ltd (Ras Al Khaimah, UAE). Subscriptions are sold on its behalf by IE Artem Polyanskiy (Georgia, Small Business, tax ID 305603825). Payments are processed by Creem as merchant of record. Terms and the privacy policy are inside @wamagentsbot.

Payments are taken by Creem, our merchant of record. Legally the subscription is sold to you by Creem (Armitage Labs OÜ, Estonia), so the charge appears under the Creem name rather than ours — this is expected. You can find the order, download the invoice or cancel the subscription at creem.io/my-orders/login using the e-mail you paid with. For anything about the service itself, write to support@magnets.tg.

By default requests go through DeepSeek infrastructure (PRC) — consent is confirmed during onboarding. If you attach your own model, chats go via your OAuth, and content is governed by that provider's API terms and privacy policy. Auxiliary services (Deepgram, OpenAI, Perplexity, Replicate, Hetzner) are listed on the subprocessors page. We keep only billing metadata and the agent container.

/delete in @wamagentsbot removes the agent along with your account. Withdrawing your data-processing consent is a separate command — /withdraw-consent withdraws the consent and then takes you through account deletion. /consent shows the current status.

Questions and bugs go to @wamagentsbot. The Service is in beta and has no guaranteed SLA (Terms §4.1); we answer complaints within 30 days and data requests within one month.