> ## 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.

# The analyst

> Ask in plain language, get an answer built from the live tape, not from model memory

The Noxint chat is an AI market analyst for crypto and US stocks, and it runs
on one rule: the model's memory is not a source. Every answer is grounded in
data pulled live at ask-time. If the data is not there, the analyst says so
instead of improvising a number.

Chat is also the proof of the engine. The same data layer that will serve
agents over the API answers you here, under the same freshness clocks and the
same refusal rules. A number you read in chat obeys the same contract as a
number an agent fetches.

## How it works

When your question needs a number, the analyst calls a live data tool to fetch
it. You watch it work through plain-language activity labels like "Pulling the
live tape" or "Reading SEC-verified financials", so you can see exactly which
data is being pulled for which asset. The final answer is built from those
reads, nothing else.

The analyst reaches across the whole picture:

* Live crypto prices and cross-venue consensus
* Derivatives positioning: funding rates and open interest
* Order flow and taker activity
* Liquidations and cluster mapping
* Technical indicators and market structure
* On-chain token screening and contract safety
* US equity financials exactly as filed, plus valuation and price history
* SEC filings, insider (Form 4) activity, and institutional (13F) ownership
* News flow, prediction-market odds, the social pulse, and the macro regime
* Web search, for narratives the data tools do not carry

## Personas

Three analysts, one data layer. **Quant** is the default: balanced, leads with
synthesis. **Degen** hunts momentum and setups and talks like it. **Contrarian**
looks for crowding and the fade. The persona changes the framing and the
emphasis, never the data. Nobody gets to make up a different tape.

## Live reasoning trace

A collapsible panel streams the analyst's thinking while the answer generates.
It exists only while the answer is live: never stored, never replayed, gone
after a reload. You get to watch the reasoning without it becoming a record.

## Working with answers

Attach images to a message: JPEG, PNG, WebP, GIF, or AVIF, up to three per
message and 5 MB each. Copy any answer. Edit any question you asked to
regenerate its answer. Startup suggestions get you moving, follow-up chips
propose the next question after each answer, and your threads live in a
paginated sidebar history.

<Note>
  A Telegram surface is built and will be enabled after launch.
</Note>