How yuask works.

Overview

yuask is an independent AI workspace built on uncensored models. One account works in every part of it:

• Chat: the web app. Streaming replies, adjustable thinking, saved chats encrypted on your device, temporary chats and export.
• CLI: the terminal app. A chat and a coding agent that reads, edits and runs commands in your project.
• API: an OpenAI-compatible endpoint, so other coding tools can use your plan and credits.
• Yuask Code: a full coding agent you install. Coming soon.

All of them draw on the same balance. There is no email, no name and no password anywhere, and payments are prepaid in stablecoins with no automatic renewal.

Your account

Your account is a public key. You prove you hold the private half by signing a one-time message; we never receive anything secret. There are two ways to do it.

Access key

Create an account generates a new key in your browser. You are asked to save it, as text or as a file, and to confirm that you did before you continue. To sign in on another device, choose Sign in and paste the key or open the file; it is read in the browser and never uploaded.

Wallet

You can also sign in with a Solana wallet (Phantom, Solflare, Backpack) or an EVM wallet such as MetaMask. The wallet shows a sign-in message that follows the Sign-In with Ethereum and Sign-In with Solana standards, so it names this site and warns you if another site asks for it. Signing is not a transaction and costs nothing.

Sessions

Once signed in, one cookie keeps you in for 30 days. It holds a random token and nothing else. In Settings you can sign out of this browser or of every session at once, and you can see your account id, which is a prefix (key:, sol: or evm:) followed by your public address.

The chat

Open /app and write. Replies stream as they are written, with Markdown and highlighted code. The controls in the message box set the model and the thinking level for the next reply, so you can change them in the middle of a conversation.

• Saved chats are encrypted in your browser before they are stored, and sync to any device where you sign in with the same key or wallet. See Privacy and encryption.
• Temporary chats are never saved, not even encrypted. Close the tab and they are gone.
• Export saves one chat as a Markdown file, or all your chats as one JSON file, decrypted, on your device.
• Usage, from the meter in the sidebar, shows this week, your plan, your credits, the last 30 days and your purchases.

Storage for saved chats: an account that has never paid can keep 5 MB and 100 chats. After the first payment of any size, that becomes 200 MB and 5,000 chats.

Models

Three models, all without a content filter. Every plan includes all of them. GLM-5.3 Flash is the default because it gives the most replies for your money; switch to a stronger one when the task needs it.

Prices are in US dollars per million tokens. “Cached” is input the model has already seen earlier in the same conversation, which is billed at the lower rate. Replies per $1 are measured with Think on conversations of ordinary length; long chats, long replies and deeper thinking use more. “Cost” is the price of a reply compared with GLM-5.3 Flash. Every model reads up to 1M tokens, within your plan’s context window.

The models can be wrong, out of date or make things up, and they will discuss what other assistants refuse. You stay responsible for what you do with a reply; see the terms.

Thinking levels

Before answering, a model can reason privately. More thinking gives better answers to hard problems and costs more, because thinking is billed as output.

Instant

No thinking, the fastest reply. Only on models that offer it.

Think

A short think before answering. The default.

Deep think

Works through hard problems step by step.

Max think

The longest reasoning, on Plus and higher plans.

The models table shows which levels each model accepts. You can open the thinking of any reply to read it.

Plans and usage

Plans are prepaid and do not renew on their own. Each one gives you its full price in usage, at the prices in the models table, and releases it a quarter at a time each week so it lasts the whole period.

Replies a week are with GLM-5.3 Flash and Think. The Trial gives its whole $1 in its single week, about 1,250 replies, and can be bought once, by an account without a plan.

How usage is counted

• Every reply is billed by the token: what you send (your messages and the conversation so far) and what the model writes, thinking included. The exact cost of each reply is in Usage.
• A week starts on the day and hour your plan started, not on a Monday. What you do not use in a week does not carry over.
• When the week’s usage runs out, replies are paid from your credits. With no credits left, the app tells you when the week resets.
• A reply never costs more than you have: its length is capped to what your balance can pay for.

Credits

Top-up packs of $5, $10, $25, $50, $100. You get the face value in credits. Credits never expire while the account exists, have no weekly limit, and work with or without a plan. Without a plan, an account runs on the limits in the “Credits only” row under Limits.

Renewing, upgrading, downgrading

• The same plan again adds 30 days and that plan’s usage on top of what you have.
• A higher plan starts at once. What was left of the old plan becomes credits, so nothing is lost.
• A lower plan bought while a higher one is active is added as credits.

Paying

Payments are in stablecoins: today USDC and USDT on Solana. The networks and tokens you can use are always the ones shown at checkout. Open Plans & credits in the app, pick a plan or a pack, then a network and a token.

• With a connected wallet, confirm the transfer in the wallet and the app does the rest.
• From any other wallet or exchange, scan the QR code or copy the address, and send exactly the amount shown.

Each checkout has its own amount, at most a few cents above the price (for example 25.0137 USDC for Plus). Those last digits are how we recognise your payment on the chain, without a memo and without linking it to anything else, so it also works from exchanges that cannot add one. Your plan or credits arrive as soon as the transfer is confirmed.

• A checkout stays open for 60 minutes. A payment that arrives after that is still matched for 48 hours.
• Network fees are paid by your wallet and are not part of the price.
• Payments are final. If you sent a different amount, keep the transaction id and contact us.
• Transfers on public blockchains can be seen by anyone, including the wallet you paid from.

Redeem codes

A code can give credits or a plan. Enter it in Plans & credits, or open a link of the form yuask.ai/app#redeem=CODE: the app offers it once you are signed in, and nothing is applied until you accept. Each account can use a given code once. A code can have a limited number of uses, an end date, or be meant only for accounts that have paid before.

Privacy and encryption

The short version: we do not know who you are, we cannot read your saved chats, and we do not keep what you send to the model. The full detail is in the privacy notice; here is how it works.

Saved chats

Your browser derives an encryption key on your device, with HKDF-SHA-256, either from your access key or from a wallet signature of a fixed message that names this site (again, not a transaction). The key is kept in the browser as a non-exportable key and never sent to us. Every chat is encrypted with AES-256-GCM before it leaves the page, and our server only ever stores the ciphertext. Neither our staff nor our admin tools can decrypt it.

On a new device, signing in with the same key, or signing the same message with the same wallet, recreates the same encryption key, and your chats open again.

Replies

To answer, the model has to see your conversation. It travels over TLS to our server, which passes it to the machines that run the model and streams the reply back to you. This step is not end-to-end encrypted. Our server writes no message content to disk or to logs; per account and per day we keep only how many replies and tokens were used and what they cost. The model’s infrastructure does not train on your data.

What is never collected

Email, name, phone number, analytics, ad trackers. Your IP address is used briefly in memory to limit abuse and is not stored in our database.

The CLI

The Yuask CLI brings the chat and a coding agent to your terminal, on the same plan and credits. It needs Node.js 20 or newer.

Signing in works like a TV app. yuask login prints a short code; open yuask.ai/device while signed in on the web, enter the code, and the terminal is connected. It receives its own token, valid for a year, which you can revoke from Settings → Devices at any time. We store only a hash of it.

yuask

Chat in the terminal.

yuask "prompt"

One reply, then exit. Fits in pipes, hooks and CI.

yuask code [task]

The agent: lists, reads, searches and edits files and runs commands in the current project. Writes and commands wait for your yes.

yuask omp [task]

The full omp harness on your account: plan mode, subagents, language-server tools, resumable sessions and MCP.

yuask usage

Your balance and the last 30 days.

yuask models

The models you can pick with -m.

yuask logout

Sign this machine out.

Flags: -m model, -e thinking level, -t show thinking, -y approve everything for this run. The CLI works on your files locally and sends only what the model needs: the task, the files it reads and the output of the commands it runs. More on the CLI page.

The API

yuask speaks the OpenAI Chat Completions format at https://yuask.ai/v1. Point any tool that accepts a custom OpenAI base URL at it (Cline, Continue, Aider, your own scripts) and it spends your plan and credits exactly like the app.

Authentication

The API key is the token of a connected terminal. Run yuask login once; the token is saved in ~/.config/yuask/credentials.json (on Windows, %APPDATA%\yuask\credentials.json) and starts with abl_cli_. Send it as a bearer token. It is a credential: keep it private, and revoke the terminal in Settings if it leaks.

$ export YUASK_API_KEY=$(jq -r .token ~/.config/yuask/credentials.json)

$ curl https://yuask.ai/v1/chat/completions \
    -H "Authorization: Bearer $YUASK_API_KEY" \
    -H "Content-Type: application/json" \
    -d '{
      "model": "glm-flash",
      "reasoning_effort": "low",
      "stream": true,
      "messages": [{ "role": "user", "content": "Explain HKDF in two lines." }]
    }'

Endpoints

GET /v1/models

The models, with their context window and longest reply.

POST /v1/chat/completions

A reply, streamed as server-sent events or as one JSON response.

Request fields

• model: one of glm-flash, deepseek-flash, glm. Anything else uses GLM-5.3 Flash.
• messages: system, developer, user, assistant and tool turns. Content is text; other parts, such as images, are ignored.
• reasoning_effort: none, low, high or max (Instant, Think, Deep think, Max think). The default is low. A level the model or your plan does not offer is refused.
• tools: up to 40 function tools. Tool calls come back in the standard tool_calls format.
• stream and stream_options.include_usage: streamed deltas carry content, reasoning_content (the thinking) and tool_calls; the last chunk reports token usage when asked.

Requests through the API get a 1M-token context window, replies of up to 128k tokens, 120 requests a minute and 6 at once, whatever your plan, because agents make many short calls on one long transcript. Billing is the same as in the chat.

Limits and errors

Limits in the chat and the CLI depend on the plan you have.

When a conversation is longer than the context window, the oldest messages are left out of what the model sees. The CLI agent gets twice the messages a minute, since one task makes several requests.

Errors come back as JSON with a readable message and a short code: {"error": "…", "code": "weekly"}.

401 auth

Not signed in, or the token was revoked.

402 weekly

This week’s usage is spent and there are no credits. The message says when it resets.

402 funds

The balance is empty, or too low for a conversation this long.

403 plan

That thinking level is part of a higher plan.

400 effort

That model does not offer that thinking level.

413 context

The last message alone is longer than your plan’s context window.

413 quota

The storage for saved chats is full.

429 rate

Too many messages in a minute. Wait a moment.

429 busy

Your plan’s replies at once are all still being written.

Your data

• Export any chat, or all of them, from the app at any time.
• Delete all chats or delete the account in Settings. Both ask you to type DELETE first. Deleting the account removes the account, its sessions, connected terminals, encrypted chats and remaining balance.
• Payment records are kept for accounting without your account id. If the account used its Trial, a one-way hash of its id is kept so the Trial is not offered to it again.

What comes next

• Yuask Code: a full coding agent you install. It plans, edits across your repository and runs your tests, with the same models, account and privacy. Read more.
• VS Code extension: chat, edits and code review inside your editor.

Follow @YuaskAI for news.