zkTerminal

zkTerminal in five minutes

Written 2026-11-06 for someone who has never seen it. Plain words, no promises

beyond what runs today. Where something is not live yet, it says so.

What it is

zkTerminal is a place to talk to an AI and to run a small AI team for your

company, with your material kept private. You sign in with an email link or a

passkey. Behind it sits Zynapse, the gateway at api.zkagi.ai, which does the

model calls, the metering and the receipts. You never paste a model key; the

gateway knows who you are from your sign-in.

Two things happen there. You talk: an ordinary chat, private by default,

gone in 24 hours unless you ask to keep it. And you hand over missions: a

goal the system breaks into steps, runs within a budget you set, and brings back

with evidence. The team is the same idea on a timer: eight named teammates who

propose work each day and do only what you approve.

The day

public sources with timestamps, what matters in the company from your own

documents, and three decisions that need you. Every number is checked against

its source; if a check fails the page says so instead of hiding it. It is

information, never investment advice.

on to talk to one teammate directly. For bigger work, say "launch a mission:

..." and watch it run; each step is billed to you with a receipt.

message, a spend, the finish of a mission. You approve exactly the text you

see, edit it (which makes a new draft), or reject it with a note the team

remembers. Nothing is sent without your yes. Channels without credentials are

simulated, never sent.

on which lane, for how much so far; who is waiting at the Door; today's spend

against the daily cap. Start, pause and the cap are on this page only, never a

chat command, so an AI holding your sign-in cannot switch itself on.

What "private" means here, in plain words

machine proves what code it runs before any text is sent, and the gateway

refuses to send if the proof fails. The other lane, "fast", is a provider with

zero data retention by contract and no hardware proof. You pick per call; the

default is sealed.

values are replaced with placeholders; the model sees the placeholders; the

real values are put back in the answer on the way to you. A company workspace

can switch scrubbing off for itself.

anyone can verify at api.zkagi.ai/r/<id>. A receipt holds amounts, counts and

hashes, never the text.

server stores only ciphertext; we cannot open it, and a lost passkey without

the recovery code means lost data. That is the trade.

one, ask in chat ("keep this"); the assistant stores it in your vault with the

`keep` tool. Missions keep their finished work for you until you delete it.

Logs hold no content.

How a company sets up (the five steps, at /id/start)

1. Sign in. One email link or passkey; the first sign-in creates the account.

2. Make your key. A passkey at /id/vault, plus a recovery code.

3. Add credit. $10 once to get started, then pay as you go; the last dollar

works like the first and credit never expires. Card (Stripe's page) or the

ZKAGI token.

4. Your company workspace. One account the whole company shares: one Brain,

one team, one balance, with membership by email domain (everyone @yourcompany

lands in it). The operator sets it up once; until then you work in your

personal account.

5. Give the Brain its first document, then switch the team on. The Brain is

your company's documents, encrypted per workspace and searched locally; the

team works only from it. The team is off until you switch it on at the

office, with a daily cap (default $3) and a limit on work in flight.

How the team works for you

Each teammate has one job, a budget per task, a context rule (company material

and recent outcomes, never your chat history) and a definition of done.

documents and what is still unknown.

at the Door.

cap the score whatever the model says.

reports whether they name you, plus trend data. Needs your website and the

operator's tool keys.

in a sealed sandbox, no network, and reports every number as printed.

spend that waits at the Door. The video studio is not deployed; Kai's

briefs run, renders do not, until the operator switches a studio on.

Rules that are code, not prompts: autonomy is off until you switch it on; a

daily cap and a cap on work in flight stop new work; nothing external happens

without a Door approval; teammates have no browser, no send, no spend; every

unit of work is a mission with a budget and receipts; pause stops new work at

once and running work finishes within its budget. "Done" means a reviewer

passed it; otherwise the status says why not.

What it costs

Pay as you go, metered per call at the gateway's published rates, with a receipt

for each. $10 of credit gets an account started; an optional $100 plan gives 30

days and the same value as credit. The team's daily cap is yours to set between

$0.50 and $50; a mission's budget is a hard cap between $0.05 and $25 (default

$2). A Morning costs a few cents. Frontier models from closed providers are a

separate paid lane, switched on per account by the operator and never used by

default; the private lanes are open models.

Live and not live, honestly (2026-11-06)

Live: sign-in, passkey vault, credit, chat on the sealed and fast lanes, the

Brain, missions, the Morning, the Door, the office with its live stream, the team

timer. Needs a deploy from this branch: the team as @-mentionable agents in chat

(seeded once on the server), the top bar, the start page. Not deployed: the

video studio's renders. Not built: scheduled missions beyond the team's daily

rhythm, posting to channels without credentials (they are simulated).