Code with the model you choose.

Cognia is a chat workspace for programming. Pick a model, ask your question, and pay for the tokens that reply used. Nothing to install and no repository to connect.

Start chatting See models and prices

Choose the model for each chat
Cost shown under every reply
Agent Keys for your own tools
No GitHub, clone, or IDE required

Your chosen modelExample

Why does total come back as NaN for an empty cart? cart.ts

reduce has no initial value, so an empty array throws and the fallback path divides by zero. Give it a starting value:

TypeScriptCopy code

const total = items.reduce(
  (sum, item) => sum + item.price * item.qty,
  0,
);

412 in, 188 out$0.0031example figures

Ask about code, paste an error

Pick the model, see its price first

The catalog comes from the upstream provider and syncs on a schedule. Each model shows what you pay per 1M input and output tokens. Cognia routes to third-party models and does not train its own.

All models

Model	Input, USD per 1M tokens	Output, USD per 1M tokens
ag/claude-opus-4-6-thinking ag/claude-opus-4-6-thinking	$0.6	$3.00
ag/claude-sonnet-4-6 ag/claude-sonnet-4-6	$0.3528	$1.76
cb/claude-opus-4.6 cb/claude-opus-4.6	$0.54	$2.70
cb/claude-opus-4.7-1m cb/claude-opus-4.7-1m	$0.54	$2.70
cb/claude-opus-5 cb/claude-opus-5	$0.48	$2.40
cc/claude-fable-5 cc/claude-fable-5	$5.40	$27.00

Prices last synced Fri, 02 Oct 2026 14:40:13 GMT. 2 models are paused because a current price is not available.

Pay as you go, to the token

There is no subscription. You add funds to a balance and each request draws from it.

Held before the request: The most a reply could cost is set aside first: your prompt, the chat history sent with it, and the output limit. Two requests can never spend the same funds.
Charged for actual usage: When the reply ends, the charge is calculated from the token counts the provider reports, at the price shown when you sent it.
The rest is released: Whatever was held but not used returns to your balance right away. A request that fails before the model starts costs nothing.
Stopping is not free, and not full price: If you stop a reply, you pay for what was generated up to that point. When the provider gives no final count, the figure is an estimate and is labelled as one.

How billing works in detail

One requestExample figures

Held while the reply runs$0.0460

Charged: 1,240 input and 356 output tokens$0.0083

Released back to your balance$0.0377

Every charge is a ledger entry you can see under Usage and Billing.

Agent Keys for your own agents and apps

Create a key and call the same models over an OpenAI-compatible API. Requests use your balance and follow the same billing as chat.

The secret is shown once and stored only as a keyed hash
Set a spend limit and a request limit per key
See every request a key made, with tokens and cost
Revoke a key and new requests stop immediately

Create an Agent Key Read the API docs

curl https://usecognia.xyz/v1/chat/completions \
  -H "Authorization: Bearer $COGNIA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "ag/claude-opus-4-6-thinking",
    "stream": true,
    "messages": [
      {"role": "user", "content": "Write a debounce helper in TypeScript."}
    ]
  }'

Three steps to your first reply

1
Sign in
Use an email link or a wallet signature. There is no password to manage.
2
Add funds
Send USDG on Robinhood Chain from a linked wallet. It is credited to your balance after onchain confirmation.
3
Pick a model and chat
Ask, debug, explain, generate, or refactor. Your chats are saved to your account until you delete them.

What to expect

Does Cognia run my code?: No. Nothing you paste, attach, or receive is executed. Replies are text for you to review and run yourself.
Whose models are these?: Third-party models reached through an upstream provider. Cognia does not have a foundation model of its own.
Do I need GitHub or an IDE?: No. The workspace is a web page. You can paste code or attach text and source files to a message.
Why does a long chat cost more per message?: Each request sends the earlier messages again so the model has context, and those tokens count as input. Starting a new chat resets that.

Open a chat and try it

Start chatting