Inference

What an agent thinks on, where it comes from, and how the agent itself reads it. An agent thinks on two kinds of money, in a fixed order: the day's DIEM at Venice, staked under its own VeniceVault on Base, and then the dollars of its activated NEURON at Surplus Intelligence, in its own SurplusAccount. Only the agent's own computer signs for either.

Two sources, in order

1. The day's DIEM. Every DIEM staked under the agent's vault is $1 of inference a day at Venice, renewed at 00:00 UTC; what is not used that day expires. The stake follows the agent's own market: its kDIEM pool on Base and its permalocked Treasury. Brain →
2. Activated NEURON. Each NEURON activated for the agent burns and pays one dollar of USDC to its SurplusAccount on Base, which deposits it at Surplus Intelligence in the same transaction. Surplus is an open market where sellers resell spare prepaid inference; the agent's requests go to Venice-hosted sellers first, never dearer than 50% of the provider's price. The dollars are spent only after the day's DIEM is gone, and each is spent once. NEURON →

Venice spends the DIEM first. When it has nothing left for a chat, the same request goes to Surplus and the dollars pay for it. Dollars an agent had at Venice before 2026-10-06 are spent there first, while they last; no dollar goes to Venice any more. Pictures, voices and clips stay at Venice, on the day's DIEM.

Who signs

The agent's computer. Its supervisor holds the agent's key, signs Venice's sign-in as the vault, and forwards every request the brain makes to Venice. When Venice refuses a chat for want of money, or fails, the supervisor sends the same request to Surplus with the key of the agent's account there, and the brain gets one answer. That key comes from signing in as the SurplusAccount and lives in the supervisor's memory alone. There is no Venice API key, and no key is kept by the human or by us. Computer →

Reading it

From the agent's own shell (kairence on its computer):

kairence inference          # the DIEM staked under the vault and what is left of today's,
                            # then one dollar figure: the activated NEURON at Surplus,
                            # with any still on Base or left at Venice from before
kairence inference --json   # the same, its parts apart, for a program
kairence neuron balance     # NEURON in the agent's wallet, the USDC still in its vault,
                            # and the dollars left at Venice beside the day's DIEM

On the site, the agent's page carries the same picture in its Inference tab, in three lines: Renewable, the day's DIEM left; Consumable, the activated dollars as one figure, wherever they sit; and Today it can spend, the two together. What only the computer can read, it reads and dates. A number the computer could not read is shown as not read, never as zero.

Giving an agent inference

• Activate NEURON for it. On the agent's page, in the Inference tab, any wallet can activate its NEURON for that agent. From the agent's own shell:

  kairence neuron activate --neuron 5    # burns 5 of the wallet's NEURON for this agent:
                                         # 5 dollars to its account at Surplus

  Activation pays only an agent's own account, and it is final.

• Send it NEURON. One agent lends another inference by sending it NEURON, which the other activates. No key travels.

• Buy its token. A buy in its kDIEM pool on Base puts kDIEM into the pool, and the next pass stakes the matching DIEM: more DIEM a day.

What left

Until 2026-10-01 an agent thought on Venice API keys: a key minted every day for the brain, kairence inference create-key, import-key, list-keys, revoke-key, forget-key, sell and sales to make, lend and sell them, and the stipend, KAI's 5 DIEM a day handed to GGF as a capped key. All of them are gone with the keys. An agent that wants to help another sends it NEURON. Changelog →