Inference for your agents.

Stake $QUOTUM for one of a hundred seats. A capped Venice key for inference each session.

The bell is at 20:00 UTC. What the keys leave unspent buys $QUOTUM and burns it.

Pay as you go is open by invite for prepaid project keys.

$1,040.86Spent on inference $4,640.40Burned at bells123.24M QUOTUM

Burned runs ahead of inference because seats use only part of their daily caps. QIP-5 backing is counted separately. Through bell 39, 3 Nov 2026. Saved, read 3 Nov at 20:47 UTC. Latest bell tx

Briareus and a hundred hands holding the quotum bell

Connect your agent

Paste this into your coding agent. It reads our docs and wires your seat key. Without a seat, use a Pay as you go project key.

Connect this project to my quotum seat.

Read https://quotum.org/llms.txt and https://quotum.org/AGENTS.md.
Set this project's client base URL to https://api.venice.ai/api/v1. Keep your own agent backend.
Read the key from QUOTUM_SEAT_KEY. Never print or commit it. No key yet: open quotum.org/seat, Other apps.
Pick a private model from quotum.org/models.
Send one small test call. Open See the call land on Your seat to inspect its cost and observation time.

Never ask for my seed or private key, sign for me, or move tokens.

Claude Code

In Claude Code, connect this project's client using the shared instructions above.

Codex

In Codex, connect this project's client using the shared instructions above.

Hermes

hermes plugins install qvotvm/hermes-quotum --enableUse QUOTUM_SEAT_KEY, or a Hermes key from quotum.org/seat, Other apps. After the test call, run hermes usage.

Cursor

In Cursor, connect this project's client using the shared instructions above.

It will

Set one env var and a base URL
Use a private model
Send one small test call

It will not

See your wallet or seed
Sign or move tokens
Swap its own model backend

For agents: llms.txt, seat guide, machine-readable guide, bell and live state.

What quotum does not do

Pay holders dividends or yield.
Let a seat key spend past the seat allowance.
Carry unused regular allowance into the next session.
Store prompt or reply text on the standard seat console server.
Independently verify Venice's prompt retention policy.

Seat keys, cap rules and questions.

A hundred seats.

Reading of 100 windows are lit. Each lit window is a seat with a key; its ink is what that key has spent against its cap allowance. What the keys leave unspent burns at the 20:00 UTC bell.

Enter a window The house, live

Reading public activity. No prompts.

Select a window to inspect its public use.

Open the houseInspect a window

A day of quotum

The keys spend. The call seals. At 20:00 UTC the bell counts what is left, and a new session opens.

Explore the day in UTC 12:00

The keys are spending

Each recorded request leaves a little less for the bell.

Reading saved observations

Inspect the call. Call deadlines follow the NYSE open, 13:30 UTC until 2 Nov, then 14:30 UTC.

A place for your agent

Stake 2,500,000 QUOTUM, returned when you leave. The admitted key cap is Loading cap, up to $50 per session.

Reading seat availability

One seat per wallet. Or pay with ETH through NVDA to QUOTUM.

Take a seat

One key, put it to work

Agent setup ↗

Connect

Bring your own agent

Use your seat key in Hermes, OpenClaw, Goose or Open WebUI.

Converse

Open the console

Chat on the seat’s own key with private and anonymized models.

Build

Call from your code

Use an OpenAI-compatible client, with Venice as the endpoint.

Day keys

Day keys are off

Off since 30 October 2026. Use your seat’s own key for its allowance.

Inspect

Keep the receipt

See tokens in and dollars out, without a prompt in the receipt.

Let the bell count

At 20:00 UTC, unspent cap buys QUOTUM on the pair and burns it.

What the keys leave unspent burns

Every day at 20:00 UTC, weekends included. Unspent cap buys QUOTUM on the pair and burns it. The seat keeps its key and receives no tokens.

If the price guard refuses the buy, the unspent amount stays for a later bell.

Every bell, with its transaction

Replay uses a completed receipt. It makes no new burn.

Past bells, as receipts

Full history, with transactions

Every bell on the usage page

Bell, UTC	Seats	Inference, USD	Burned	Supply	Transactions

Shipped, with a date

All changes ↗

Date	Release	What changed	Type
Reading dated releases

Before you call

What funds the keys?

Trading tax funds the session budget. The public contract balance is being read. Session funding and agent backing have separate conditions; see the full conditions in the published proposals. Inspect the published proposals.

How much can the key spend?

Up to $50 per session, with the cap recalculated at each open. The admitted cap this session is being read. This is a public cap, not your signed-in balance. You can unstake whenever you like; the key is deleted and your stake returns.

How does pricing work?

A seat uses its Venice key within the session cap. Pay as you go is open by invite for prepaid project keys; check /api/payg/v1/status before sending USDG. x402 per-request intake stays closed. See the seat-key quickstart for seat access.

Who can see my prompt?

Venice labels these models private and states that it keeps no prompt or reply; that is the provider's policy, not something quotum can verify. For anonymized models: The upstream provider sees the prompt, without who sent it. Receipts keep tokens and dollars, not your prompt. Read what quotum stores.

The mechanics and the measure

A seat's open cap is the smallest of $50, its share of the effective opening input, and 0.25% of the supply at the previous session's average price. The effective input is measured net trading tax plus any reserve uplift to the $10 floor voted by holders (QIP-1), paid from the sink reserve. The admitted key cap can differ from the open cap.

Nothing is minted and there is no second token. Every burn uses the token's own burn(), so the supply only goes down.

Read the mechanics and the measure

The key from a terminal

from openai import OpenAI
client = OpenAI(
    base_url="https://api.venice.ai/api/v1",
    api_key="your seat key"
)
reply = client.chat.completions.create(
    model="deepseek-v4-pro",
    messages=[{"role": "user", "content": "Hello"}]
)

The key is created after your stake confirms and deleted when you leave. Use it in an OpenAI-compatible client.

Agent setup

Private and anonymized models

With private models Venice states that it keeps no prompt or reply. With anonymized models the provider sees the prompt without your identity. Tools and web reads leave the box. The receipt records tokens in and dollars out, never your prompt.

Today's cap buys about being read output tokens of deepseek-v4-pro. Input tokens also cost money.

Each model's privacy and list price

Payments to Venice and the public balance

Forwarded, UTC	NVDA	USD at feed	Tx

Available for sessions: being read, enough for being read sessions for today's being read seats at the current cap.

One custodial step, every payment posted. The tax lands in a contract with no withdraw path. It forwards to one published payment address, which converts it and pays the inference provider. There is no trustless way to pay the provider yet.