thetadatadx (Python)

The Python SDK for ThetaData market data. Pull US stock, option, index, and rate data three ways — point-in-time history, real-time streaming, and whole-universe flat files — all from a single authenticated client. Connects straight to ThetaData; nothing to install and run locally, no local proxy.

Features

• Complete coverage: stocks, options, indices, and rates across 64 typed endpoints.
• Three access modes, one client — point-in-time history, real-time streaming, and bulk flat-file downloads.
• DataFrames built in — every result chains straight to Polars, pandas, or Arrow over a zero-copy boundary.
• Typed all the way down — every tick is a typed object with attribute access and IDE completion, not a dict.
• No terminal to run — a direct connection to ThetaData; nothing to install and babysit locally.

Install

pip install thetadatadx-py

pip install thetadatadx-py[polars]   # Polars DataFrames
pip install thetadatadx-py[pandas]   # pandas DataFrames (Arrow-backed)
pip install thetadatadx-py[arrow]    # raw pyarrow.Table
pip install thetadatadx-py[all]      # every optional adapter

Binary wheels ship for Linux, macOS, and Windows and require no Rust toolchain. Wheels use CPython's stable ABI (abi3), so one wheel per platform covers Python 3.12 and up; the free-threaded (3.14t) build runs with the GIL disabled and is selected automatically by pip.

Quick start

from thetadatadx import Client

# Pass your API key directly. Use market_data_type="STAGE" to target staging.
client = Client(api_key="td1_...")

# First-order Greeks for every strike on SPY's 2026-06-18 expiry, as of 2024-03-15
greeks = client.market_data.option_history_greeks_first_order("SPY", "20260618", date="20240315")

df = greeks.to_polars()
print(df.select(["strike", "right", "delta", "theta", "vega"]).head())

Other ways to construct the client:

from thetadatadx import Client, Credentials, Config

# API key from the THETADATA_API_KEY environment variable, or from a .env file
client = Client.from_env()
client = Client.from_dotenv(".env")

# Email and password, inline
client = Client(email="you@example.com", password="your_password")

# Full control: build a typed Credentials + Config (custom hosts, timeouts)
client = Client(Credentials.from_file("creds.txt"), Config.production())

Every market-data method returns a typed list — iterate it, index it, or convert it to a dataframe:

eod = client.market_data.stock_history_eod("AAPL", "20240101", "20240301")
for tick in eod:
    print(f"{tick.date}: O={tick.open:.2f} H={tick.high:.2f} "
          f"L={tick.low:.2f} C={tick.close:.2f} V={tick.volume}")

bars = client.market_data.stock_history_ohlc("AAPL", date="20240315", interval="1m")   # 1-minute bars
exps = client.market_data.option_list_expirations("SPY")
strikes = client.market_data.option_list_strikes("SPY", exps[0])

DataFrames

Every result converts directly to a dataframe — no row-by-row iteration:

greeks.to_polars()   # polars.DataFrame
greeks.to_pandas()   # pandas.DataFrame   (pip install thetadatadx-py[pandas])
greeks.to_arrow()    # pyarrow.Table      (zero-copy)
greeks.to_list()     # list[GreeksTick]

The .to_arrow() terminal hands the underlying Arrow buffers to pyarrow over the Arrow C Data Interface — zero-copy at the boundary — so the table plugs straight into DuckDB, polars, cuDF, or Arrow Flight:

import duckdb

table = client.market_data.stock_history_eod("AAPL", "20240101", "20240301").to_arrow()
con = duckdb.connect()
con.register("eod", table)                 # zero-copy into DuckDB
con.sql("SELECT AVG(close) FROM eod").show()

List endpoints (stock_list_symbols, option_list_expirations, …) return a StringList with the same terminals; the single column is named by the endpoint (symbol, expiration, …). Empty results still convert to a zero-row frame with the full typed schema.

For multi-day backfills, stream the response instead of buffering it. Every market-data builder exposes .stream(handler) / .stream_async(handler) alongside the buffered .list() / .list_async() terminals; the handler is called once per chunk with a typed list, and the previous chunk is freed before the next is fetched, so peak memory stays flat regardless of total size:

def on_chunk(ticks):
    for t in ticks:
        ...   # write to Parquet, push to a bus, accumulate stats

(client.market_data.option_history_quote_builder("QQQ", "20260515").date("20260515")
    .interval("tick")
    .strike_range(5)
    .stream(on_chunk))

Streaming

Real-time quotes and trades flow through the same client. Register a callback and match on typed event classes — Trade, Quote, Ohlcvc, OpenInterest for market data, plus one typed class per lifecycle event (Connected, LoginSuccess, Disconnected, Reconnecting, …):

import time
from thetadatadx import Contract, Quote, Trade

def on_event(event):
    match event:
        case Trade(price=px, size=sz, exchange=ex, ms_of_day=ms, sequence=seq, condition=cond, contract=c):
            print(
                f"{c.symbol} {c.expiration} {c.strike} {c.right} trade price={px:.2f} size={sz} "
                f"exchange={ex} ms_of_day={ms} sequence={seq} condition={cond}"
            )
        case Quote(bid=b, ask=a, bid_size=bs, ask_size=asz, bid_exchange=bx, ask_exchange=ax, ms_of_day=ms, contract=c):
            print(
                f"{c.symbol} {c.expiration} {c.strike} {c.right} quote bid={b:.2f} ask={a:.2f} "
                f"bid_size={bs} ask_size={asz} bid_exchange={bx} "
                f"ask_exchange={ax} ms_of_day={ms}"
            )

spy_call = Contract.option("SPY", expiration="20261218", strike="550", right="C")

with client.streaming(on_event) as session:
    session.subscribe_many([spy_call.quote(), spy_call.trade()])
    time.sleep(60)   # park the main thread while events flow into on_event

Build subscriptions with the fluent Contract API and pass them — one at a time or in bulk — to subscribe / subscribe_many. Every subscription is the same typed value, so quotes, trades, open interest, and market value across contracts mix freely in one list:

from thetadatadx import Contract, SecType

stock  = Contract.stock("AAPL")
option = Contract.option("SPY", expiration="20261218", strike="550", right="C")

with client.streaming(on_event) as session:
    session.subscribe(stock.quote())
    session.subscribe_many([option.quote(), option.trade(), option.open_interest(), option.market_value()])

The option constructor is Contract.option(symbol, *, expiration, strike, right) — the leg parameters are keyword-only, so the call site always reads expiration=…, strike=…, right=… and never depends on argument order. Pair it with Contract.stock(symbol) for equities.

Or take a whole-market feed — every option trade across the universe, no per-contract setup. The full-trade feed sends a quote and an OHLC bar before each trade, so add an Ohlcvc case to the callback to handle the bars:

from thetadatadx import Ohlcvc

def on_full_trade(event):
    match event:
        case Ohlcvc(open=o, high=h, low=lo, close=cl, volume=v, contract=c):
            print(
                f"{c.symbol} {c.expiration} {c.strike:g} {c.right} bar "
                f"o={o:.2f} h={h:.2f} l={lo:.2f} c={cl:.2f} volume={v}"
            )
        case _:
            on_event(event)   # reuse the quote/trade handling above

with client.streaming(on_full_trade) as session:
    session.subscribe(SecType.OPTION.full_trades())
    time.sleep(60)   # the callback runs on the streaming thread — keep it fast

Watch feed health from the main thread without touching the callback. The session resolves the client's observability getters directly: millis_since_last_event() is the staleness clock (a steadily growing value is the earliest sign of a wedged link), ring_occupancy() against ring_capacity() shows how close the consumer is to falling behind, and dropped_event_count() is the cumulative tally of events shed on a full ring:

with client.streaming(on_event) as session:
    session.subscribe(SecType.OPTION.full_trades())
    while True:
        time.sleep(5)
        stale_ms = session.millis_since_last_event()   # None until the first frame
        print(
            f"stale={stale_ms}ms "
            f"ring={session.ring_occupancy()}/{session.ring_capacity()} "
            f"dropped={session.dropped_event_count()}"
        )

Prefer columns? client.stream.batches(...) is a sibling to the callback — the same subscriptions, delivered as pyarrow.RecordBatch values under a fixed schema. The reader is iterable (sync, releasing the GIL on the blocking pull) and async-iterable, and closes on context-manager exit:

# `batches(...)` starts the streaming session, so open it first, then subscribe.
with client.stream.batches(batch_size=8192) as batches:
    client.stream.subscribe(Contract.stock("AAPL").trade())
    for batch in batches:        # or: async for batch in batches
        print(batch.num_rows)

Flat files

Whole-universe daily snapshots for one (security type, request type, date) at a time. The decoded schema follows the request type, so flat-file results chain through the same DataFrame terminals as history:

rows = client.flat_files.option_trade_quote(date="20260428")
print(len(rows))
df = rows.to_polars()                       # or .to_pandas() / .to_arrow() / .to_list()

# Generic dispatcher when security type / request type come from config
oi = client.flat_files.request("OPTION", "OPEN_INTEREST", "20260428")

# Or write the raw vendor file straight to disk — no decode, no row materialise
path = client.flatfile_to_path("OPTION", "TRADE_QUOTE", "20260428",
                            "/tmp/option-trade-quote", format="csv")

The flat-file distribution serves a fixed set of datasets: option trade_quote / open_interest / eod, stock trade_quote / eod. Available flat_files.* methods: option_trade_quote, option_open_interest, option_eod, stock_trade_quote, stock_eod, plus request(sec_type, req_type, date). The generic request(...) and flatfile_to_path(...) paths reject an unserved (security, request) pair with a typed invalid-parameter error.

Endpoint coverage

64 typed endpoints across stocks, options, indices, the market calendar, and interest rates, plus real-time streaming.

Every endpoint is a method on Client. The full per-method list with signatures lives in the API reference; Config.dev() and Config.stage() target the non-production environments.

Errors

Every call raises a typed exception under a common ThetaDataError base — AuthenticationError, RateLimitError, NotFoundError, DeadlineExceededError, InvalidParameterError, and the rest — so the same cases are catchable here exactly as they are in every other binding.

Documentation

• Documentation site — getting started, API reference, streaming, flat files
• Repository, issues, contributing
• Community discussion on the ThetaData Discord

License

Licensed under the Apache License, Version 2.0.