> ## Documentation Index
> Fetch the complete documentation index at: https://docs.noxint.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# What is Noxint

> The data layer for autonomous trading: an AI analyst you can use today, built on an agent-first market data engine

The next wave of market participants is not a person staring at nine charts.
It is agents: models that read positioning, weigh the tape, and act. And those
agents have a problem nobody solved. Every market API still speaks human. Raw
venue dumps, mystery units, silent zeros where the data went stale. Feed that
to a model and you burn a six-figure token bill teaching it what a number
means, and it still cannot tell fresh from dead.

Noxint is the data layer built for that wave. One engine turns raw market
chaos across crypto and US equities into finished, machine-ready facts. Every
number arrives already ranked against its own history. Every response says
when it was observed and when it changes next. Anything that cannot be proved
is refused with a typed reason instead of faked.

You can talk to that engine today. Connect a wallet at
[noxint.ai](https://noxint.ai) and the analyst answers from the live tape.

## Three ways in

<Columns cols={3}>
  <Card title="Chat" icon="messages" href="/chat/overview">
    The analyst, live now. Wallet in, 100 trial credits, answers grounded in
    live data. No subscription, no email, no card.
  </Card>

  <Card title="API and MCP" icon="code" href="/api/quickstart">
    The agent lane. 23 operations over HTTP or MCP, self-teaching, priced per
    delivered result. Running in production behind chat; opens in beta after
    launch.
  </Card>

  <Card title="$NOXINT" icon="coins" href="/token/overview">
    The token on Robinhood Chain. Verified holding opens a daily flow of chat
    credits. Utility you can meter, not yield you should doubt.
  </Card>
</Columns>

## The agent data problem, solved

An agent does not need another raw payload. It needs a finished market picture
that fits in a context window. That is a formatting problem, an enrichment
problem, and an honesty problem, and the engine solves all three:

* **Context beside the value.** A funding rate arrives as
  `{"value": 0.006, "pctl": 87}` with its typical band. The model does not
  guess whether a number is extreme; the data already says.
* **Freshness is part of the fact.** `as_of` is source-event time and
  `next_update_at` says when to look again. No polling blind, no trusting
  cache time.
* **Refusal before fabrication.** Stale, incomparable, or unsupported values
  are omitted under a typed reason. Nothing is silently zero-filled, so an
  agent never trades on a placeholder.
* **Compression that cuts your token bill.** In a measured BTC positioning
  workload, four Noxint feeds cost about 3,010 input tokens against roughly
  204,100 for the mapped raw venue payloads. Same information, about 1/68th
  of the window.
* **Filed values stay exact.** Equity fundamentals come from filed statements
  as exact decimal strings with explicit units, not reconstructed floats.

The full spec is in the [data contract](/data/contract). It is the most
important page on this site.

## Economics built to survive

Most token models die of bad arithmetic, not bad intentions. Ours is metered
like infrastructure: every billable action reserves its maximum cost before
work starts and settles to actual cost after delivery. Duplicates are detected
durably and charged once. All sponsored usage (the trial and Holder flow)
lives under a hard monthly solvency cap, so free service can never outrun the
service's ability to pay for it. The math is public in
[Tokenomics](/token/tokenomics).

<Note>
  \$NOXINT has not launched yet, and everything gated on verified holding stays
  off until it is live. The API and MCP run in production behind chat today;
  public access opens in beta after launch. [Launch status](/token/status) is
  the honest ledger of what is on right now.
</Note>