> For the complete documentation index, see [llms.txt](https://docs.obside.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.obside.com/data/sources.md).

# Data & sources

Good answers need good data. This page explains where Obside's data comes from, what our AI (chat and agents) can actually see, and how we keep it fresh and honest.

## The short version

* **Market data** for stocks, ETFs, indices, forex and crypto — real-time and historical — from institutional-grade licensed providers and exchange-direct feeds.
* **A proprietary real-time data layer** that continuously ingests **130+ specialist feeds** beyond prices: macro, positioning, regulation, energy, crypto on-chain, sentiment, world events.
* **Official sources cited by name**: central banks, regulators and international institutions (Federal Reserve, ECB, IMF, World Bank, BIS, SEC, FINRA, CFTC, EIA…).
* **Point-in-time discipline**: backtests only see what was known at the time — no lookahead bias.
* **Transparency in chat**: ask the assistant what it used, and it will tell you.

{% hint style="info" %}
We name official and public institutions openly. Some commercial data partners are not named individually in public documentation for contractual reasons — all of them are licensed, institutional-grade providers.
{% endhint %}

## Market data (prices)

Real-time and historical prices for **stocks, ETFs, indices, forex and crypto**, sourced from licensed institutional providers and, for crypto, exchange-direct feeds. Historical depth goes back years, which is what powers charting and [backtesting](/agents/backtest-an-agent.md).

## Fundamentals & company data

* **Financial statements, earnings and estimates** — income statements, balance sheets, cash flows, earnings surprises.
* **Valuation and factor metrics** — P/E, growth, quality, momentum-style factor scores.
* **Regulatory filings** — SEC EDGAR filings, monitored continuously.
* **Positioning** — short interest (FINRA), futures positioning (CFTC Commitments of Traders), institutional holdings.

## Macro & central banks

Official series and announcements from the **Federal Reserve (FRED)**, **ECB**, **IMF**, **World Bank**, **BIS** and national central banks, plus a global **economic calendar** (rate decisions, CPI, employment reports…). Agents can react to these events the moment they are published.

## Energy & commodities

Official energy data (**EIA**), commodity prices, and fuel-price series — useful context for inflation, energy stocks and macro-driven strategies.

## Crypto-specific data

Beyond spot prices: global market metrics, **DeFi** activity, **network fees**, **derivatives data** (funding rates, open interest, options skew), and crypto **Fear & Greed**.

## News, sentiment & social

* A continuous **financial news flow**, deduplicated and classified, that agents can subscribe to.
* **Social sentiment** and **prediction-market** signals.
* **Accounts you choose to follow**: agents and the [SmartChart](/agents/smart-chart.md) can track posts from specific **X (Twitter) accounts** and **public Telegram channels**.
* **Live web search** in chat, so answers aren't limited to a training cutoff.

## World events

Structured monitoring of **geopolitical developments**, **sanctions lists**, **natural disasters**, **cyber incidents** and **public-health events** — the kind of context that moves markets but never shows up in a price feed on its own.

## Composite signals

On top of the raw feeds, Obside computes **proprietary composite signals** — normalized 0–100 scores that condense many raw series into one readable gauge: market stress, volatility regime, fear & greed, funding divergence, sanctions pressure, surprise momentum, and more. Agents can use them directly as conditions.

## Freshness & reliability

* **Multi-provider redundancy** — key concepts are covered by more than one source, with priority-based routing between them.
* **Staleness detection** — every data point carries its timestamp; the system knows when a series is out of date instead of silently serving old values.
* **Point-in-time storage** — history is stored as it was known at each moment. When a backtest asks for "the ranking on March 3rd", it gets what was actually knowable on March 3rd, not today's revised version. No lookahead bias.
* **Deduplication and normalization** — the same event reported by multiple sources is stored once, in a consistent format.

## Ask the assistant

Transparency is enforced at the AI level: the assistant is designed to **never invent a source**. It only cites data it actually retrieved, and if it doesn't have something, it says so. If you want to know where a specific answer came from, just ask — *"what sources did you use?"* — directly in the chat.