kalshi-python-sdk: what it does, what it costs, where it stops

What it does

Sync and async clients share one transport. Orders go through the V2 event-market family (create_v2, amend_v2, decrease_v2, cancel_v2, cancel_all_v2 plus batched variants) using Decimal prices and a required client_order_id for idempotency. Only GET, HEAD and OPTIONS retry; POST and DELETE never retry, to avoid duplicate orders.

PerpsClient and AsyncPerpsClient cover Kalshi's margin exchange (48 REST operations, 6 WebSocket channels) with separate API keys, and KlearClient covers the Self-Clearing-Member settlement API (23 operations) with bearer auth. The FIX engine speaks FIXT.1.1 and FIX50SP2 for order entry, drop copy, market data, post-trade and RFQ sessions.

• Twelve subscribe_* WebSocket methods (ticker, trade, orderbook_delta, fill, market_positions, user_orders, order_group, market_lifecycle, multivariate_lifecycle, communications, cfbenchmarks_value, cfbenchmarks_value_5hz) plus a generic escape hatch.
• Pydantic v2 models, mypy strict, py.typed shipped; optional pandas or polars extras for DataFrames.
• Record and replay transports in kalshi.testing for offline tests.
• Demo environment supported via KALSHI_DEMO=true or demo=True.

Pricing

Free under MIT. PyPI shows version 14.0.0 published September 6, 2026 and 34 releases since May 10, 2026. Python 3.12 or newer is required.

Venues and data

Kalshi only, across three surfaces: the prediction market API at api.elections.kalshi.com, the Perps API at external-api.kalshi.com, and the Klear settlement API. Contract tests are described as failing hard on query, body and WebSocket payload drift against Kalshi's published specs.

Pagination returns a Page object with a cursor, and list_all() iterates until the server returns no cursor, with an optional max_pages bound. Pages convert to DataFrames when the pandas or polars extra is installed.

Where it falls short

Six GitHub stars and a single maintainer is a thin base for something you would route live orders through. Fourteen major versions in four months signals a fast-moving API surface; the docs site carries v2 to v3 and v3 to v4 migration pages, so expect more.

Python 3.12+ excludes many existing environments. The inventory note of 107 of 108 operations is already stale against the README's 108 of 109, which is a sign of how often the spec and the package move.

Who should use it

Systematic Kalshi traders who need FIX or Perps from Python, or who want the entire REST surface typed rather than hand-rolled from the OpenAPI spec.

For a smaller, higher-starred community client with pandas and Jupyter conveniences, pykalshi is the other Python option.