Help

A personal agent with its own knowledge base. Hand it anything: writing, research, organising, work in your connected apps and on your computer. What it does best is information: it keeps watch on what you care about, reads what's new, writes briefs where every item links to its source, and keeps what matters in your knowledge base and library.

Your one ongoing chat with your agent. It carries on like talking to a person: you don't start a new one for each thing, and it remembers what you said before. Quick things are simply done. Work that runs for a while (setting up a watch, deep research, transcription, an import) shows up in the chat as a progress card you can open to see each step. Everything it made or changed is in the chat, one click away, and can be undone. You don't scroll back to find things: the chat's details have an outline by day, the things made here, and search.

You start with three: Mento takes anything you hand it, Marlo keeps watch on fields, sources and pages, and Calli looks after what you save and write. Select ＋ at the top of the sidebar and choose "Create an agent" to land in a chat with it: it asks what it should look after, then names itself and writes down its duties; or just ask Mento to set one up. You can add up to 10. Agents you add can have their duties changed and can be deleted; Mento can be renamed and given a new avatar, Marlo and Calli can only be renamed.

Yes. Choose "New group chat" under ＋, pick a few agents in the To row (up to 6), and the group is created when you send the first message. With nobody named, it first works out who this is for and the members it picks answer in turn, later ones saying nothing if they have nothing to add, and nobody answers when the message needs no answer; whoever you @ answers; with @Everyone each member answers once, in turn.

Only when you name that agent ("have Marlo look at this page", or an @ mention) or the work clearly belongs to what another agent looks after; everything else it does itself, however big. The helper gets only a brief, never your conversation, and the exchange between the two agents shows up as one line in your chat that you can open. When the result is back, the agent that asked tells you; if the helper can't do it or runs into an error, it tells you right away.

Radar is the way in: today's briefs, the watches you follow and Explore are all on that page. A watch is one thing on Radar that your agent keeps an eye on: a field (it finds the sources), specific sources (sites, podcasts, channels or newsletters you name) or page changes (pricing pages, docs, changelogs). Each watch writes briefs at the cadence you choose, and alerts tell you the moment something matches.

A knowledge base holds documents you can edit; the library holds everything else. Documents you and your agent write live in a knowledge base, and everyone starts with "My knowledge base". Pages, files and recordings you save, and the transcripts your agent produces, live in the library. You can file a library item in a knowledge base or leave it unfiled; your agent finds it either way.

A watch produces a brief that you read and others can follow. A routine does one thing for you at a set time or when something happens, such as rolling several watches into a document every Monday, or drafting a reply when a certain person emails you. The result of each run is a message in your chat, and it can also be emailed, written into a document or sent to Telegram.

Choose Connect apps at the bottom left. Apps are what your agent can use on your behalf, like Gmail, Notion and Slack. Skills are ways of doing a kind of work, like "Earnings quick read". Connect once, then just ask for it in your chat.

Its memory is a page of its own, under the account menu at the bottom left: one line each, every line showing which message it came from. You can edit or delete any line and export it all as Markdown. It notes lasting preferences you mention. It only remembers what you said yourself, and never passwords or one-off requests.

It tries not to interrupt you. Reading pages, searching, reading your documents and connected apps, anything in waper it can undo (creating or editing documents, saving items, remembering what you said), and what you asked for in the chat yourself, naming the action and who it is for, it just does, leaving a line in the chat with Undo where it can be undone. Only two things always need your approval: a payment, and sending in your name to someone you haven't mentioned. Everything else goes through the auto-review: it looks only at what you said and at the action itself, goes ahead when it is what you wanted, shows you a card only when it isn't sure, and blocks what is clearly unsafe, saying why. A single action estimated at over 600 credits also asks first, once. Choosing "Always allow" on a card lets that action on the same target go ahead from then on; you can see and remove these under Rules in settings. See When it asks first.

Each watch is written in one language: its creator's content language. The content language is what your new watches, summaries, answers and reports are written in: it follows the interface language and switches on its own when you keep talking to your agent in another language, so there's nothing to set. The language menu in the header or account menu switches the interface language at any time; when you're signed in it is saved to your account, and opening waper.ai takes you straight to that language. Watches you follow from others appear in the language they're written in; when that differs from your content language, public watches show a translation into your content language by default, marked "Translated by waper Agent" (no credits). You can switch back to the original any time, and the watch remembers your choice. Every brief of a public watch is translated into each language waper supports, and each version has its own address (the Chinese version of an English watch is at waper.ai/zh/w/…, for example); when you're not signed in, a watch page shows the original with a link to the other language. For a private watch, translation starts only when you click Translate and costs credits at the actual model cost.

Free is for trying it out: 10 followed watches, 1 new watch a month, 100 credits a day, 1 site, audio and video up to 60 minutes each, and 7 days of document history. That's enough to see what it can do for you. When you rely on it for work, that's Pro: 200 followed watches, 40,000 credits a month, 50 routines and unlimited document history, enough to follow a whole industry and make deep research and long recordings routine. Personal sits in between, for people who use it every day at a smaller scale.

Chats, deep research and page summaries use credits at the actual model cost: a typical chat turn is about 5–10 credits, and a deep research report about 12–60 (60 is set aside before it starts and settled at what it really used). Video and audio are by length: 10 credits a minute, transcript, summary and translation included. You're only charged when it works.

Daily credits first (they reset every day), then the current period's (monthly or trial), then gift credits from referrals or support (soonest-expiring first), then purchased packs. Free credits always expire; purchased credits never do.

When a friend signs up with your link and, within 30 days of signing up, verifies their email, has been registered for at least 3 days and has used waper on 2 different days (activation), you each get 600 credits (valid 60 days); you can earn this 10 times. If those aren't all met within 30 days, no activation reward is given. When a friend first subscribes, you get 2,000 (Personal) or 5,000 (Pro) credits (valid 90 days), with no limit. Accounts on the same network or mailbox and disposable addresses don't count.

Paste a YouTube, Bilibili or podcast link, or upload an audio or video file: waper turns it into a transcript and a summary, translated into your language, with a timestamp on every passage so you can listen back. It's 10 credits a minute; you see roughly how many minutes and credits before it starts, and a link someone has already processed is free to reuse.

You can try Pro free for 7 days, with no card, once per account. You start it yourself, from the pricing page or from the upgrade dialog that appears when you run out of credits or reach a paid feature. You can create up to 5 watches during the trial (following others' watches doesn't count), and it includes 1,000 trial credits shared across the whole trial. Afterwards you're back on Free and keep everything; watches created during the trial beyond the free quota pause their daily briefs (newest first) until you upgrade.

Yes. Choose "Import a Markdown folder" from the + next to "Knowledge bases" in the sidebar, and the whole folder becomes a new knowledge base with the same folder structure. Your original folder isn't touched; from then on they are two independent copies. For a folder you'd rather not upload, don't import it: choose "Work in a folder on your computer" from the + in the input box and your agent reads and edits it in place while your computer is online.

Open a watch you own and create a forwarding address in its settings, then use it to subscribe or set up auto-forwarding. Each email is split into linked items for your brief.

Email the address shown under Channels in Settings from your account email, or simply reply to your daily brief. Your agent answers by email. Mail from other addresses is ignored.

Create a key under Settings → API and MCP and add waper as an MCP server as shown there. They can then read your briefs, library and documents, save pages, write documents and create watches. See MCP.

Items you save from briefs, any web page, PDF, Word, Markdown and text files you upload, and transcribed video and audio. Each gets a summary and key points, is searchable, and your agent can read the full text to answer questions. You don't choose a place when you add something: it starts as unfiled, and when your agent is confident where it belongs it files it in a knowledge base and marks it. Change it any time.

Yes. Call /api/v1 (REST, JSON) with the same API key; the description is at /api/v1/openapi.json, and you can browse and try it in the API reference. Up to 60 requests per minute per account.

Export OPML from your reader and choose Import OPML from Radar's Add menu. Each folder can become a “specific sources” watch.

Yes. Install the desktop app on a Mac, or the command-line daemon on a server (both are coming soon and can't be installed publicly yet), and while the computer is online your agent can read and write files and run commands there, even when you ask from the web or your phone. Your files stay on your computer; what your agent reads passes through the cloud model. Reading files it just does; writing files it does directly when you asked for it in the chat, otherwise the auto-review decides, and every write can be undone; deleting moves the file into waper's own trash on that computer, so it can be undone too; running a command goes through the auto-review, which asks you only when it isn't sure (a command you chose "Always allow" for on that computer just runs); wiping a disk or your home folder is never done. It can use the whole computer by default, or you can limit each computer to a few folders. See Your computers.

Yes. Documents are Markdown: export one at a time, or export a whole knowledge base as Markdown with its folder structure. Transcripts can be exported too. Deleting your account deletes all of your data.

waper does not train models on your content or sell your data. Your questions, private watches and uploads are processed by third-party model and transcription services, as described in the privacy policy.

No. Private watches are visible only to you, and never appear in public briefs or in Explore.

Sources are official APIs, RSS and public pages; clipped pages keep summaries only, never full text; every item in a brief must come from that cycle's real fetch.

Email support@waper.ai.