What it is
Kalshi's Trade API is the exchange's own interface to its event-contract markets — REST, WebSocket and FIX, with a separate surface for the exchange's perpetual futures. The developer documentation that this page's visit link points to now serves from docs.kalshi.com. Production REST calls go to https://external-api.kalshi.com/trade-api/v2 and streaming to wss://external-api-ws.kalshi.com/trade-api/ws/v2, with a demo environment on its own hosts and its own credentials.
Two things separate it from most entries in this directory. It is the venue itself rather than a vendor reselling venue data, and large parts of it are open: series, events, markets and order books all answer without an API key.
Why a prediction-market reader cares
Every other data source in this directory describes Kalshi from the outside, and most of them only where Kalshi overlaps with something else. Odds APIs carry Kalshi as a bookmaker slug for sports markets where their coverage lines up — NFL and NBA in the case of OddsPapi, per its own developer guide — and aggregators index the markets that carry volume. Kalshi's own API is the only place that answers for the whole exchange: elections, economic indicators, weather, technology, entertainment, and the combination markets built out of them.
It is also the primary source for the things a secondary source cannot reconstruct: the full order book with the size behind each level, the settlement result and its timestamp, the market lifecycle events, and the trading surface itself. Where an odds or analytics API gives you a price, this gives you the book that produced it and the exchange's own record of how the market resolved.
Public endpoints, no key required
The documentation's market-data quick start walks through three unauthenticated calls: fetch a series, list its markets filtered by series_ticker and status=open, then pull the order book for one ticker. No authentication headers are needed for any of them, which makes the API usable for research before you have ever signed up — and makes it the cheapest way to check whether a market you care about exists and how thin it is.
Signing in, not logging in
API keys are asymmetric key pairs rather than shared secrets: you sign each request with a private key and the exchange verifies it against the registered public key. Ed25519 is the recommended type; RSA-PSS with SHA-256 remains supported for clients limited to it, including SDK versions before 3.31.0. Each request carries KALSHI-ACCESS-KEY, KALSHI-ACCESS-TIMESTAMP and KALSHI-ACCESS-SIGNATURE, where the signature is the base64 encoding of the signed concatenation of the timestamp, the HTTP method and the path — with the query string stripped. The private key is shown once at creation and cannot be retrieved afterwards, and demo credentials do not work against production.
The order book returns bids only
The order book endpoint returns yes_dollars and no_dollars bid arrays and no asks, which looks like a gap until you work through the arithmetic of a binary market: a YES bid at price X is identical to a NO ask at $1.00 − X. Every ask is therefore implied by the opposite side's best bid, and the endpoint avoids publishing the same information twice. It has a practical consequence for reading the data: the two arrays are sorted ascending, so the best bid is the last element of each array, and the spread you can actually trade is the implied ask minus the best bid — not the distance between the two highest bids.
Prices and contracts are fixed-point strings
Prices arrive as _dollars strings with up to four decimal places and contract quantities as _fp strings with up to two, so sub-cent pricing and fractional contract sizes are representable — and integer-cent fields cannot express a sub-cent price, which is why the dollar strings are the ones to read. Each market's valid prices are a grid, published as price_ranges (start, end, step bands) and labelled by price_level_structure; that grid is the source of truth and off-grid prices are rejected. Whole-cent prices are valid in every structure, and the newer tapered grids put a finer tick below $0.10 and above $0.90, where a one-cent move swings implied probability by far more than it does mid-range. Combination markets get the finest grid of all, at $0.0001. Sub-cent prices and fractional sizes can also produce balance changes finer than an account's precision, which is why the exchange applies a rounding fee with a per-order accumulator and rebates rather than simply truncating.
Throughput is the constraint, not the invoice
What the documentation rations is requests, not dollars: each call costs tokens, your tier sets the rate at which your budget refills, and your sustained request rate is budget divided by cost. Most requests cost the default of 10 tokens, and GET /account/endpoint_costs is the authoritative list of the exceptions. Read and write have independent budgets — GET traffic draws on read, while order placement, amends, cancels, order groups, the RFQ quote flow and block-trade accepts draw on write — and the split is by operation rather than protocol, so REST and FIX drain the same buckets. Tiers run from Basic at 200 read / 100 write tokens per second to Prestige at 12,000 / 9,600.
Where it sits in this directory
Three layers are worth keeping apart. OddsPapi puts Kalshi beside 350+ sportsbooks so that a Kalshi price can be judged against a sharp reference rather than in isolation. Dune dashboards and the index and analytics providers supply historical and aggregate views across venues. Polymarket's own CLOB and Gamma APIs are the sibling official venue API for the other large prediction market. Kalshi's API sits underneath all of that as the primary record for one venue: deepest on the markets nobody else carries, and the only route to execution — but single-venue by definition, so it can never tell you whether a Kalshi price is good relative to anywhere else.
Numbers that move under you
Three figures in these docs are explicitly dynamic and should be read at runtime rather than hard-coded. Endpoint token costs change, with the default of 10 the only fixed one. Tier entitlements are grants: volume-earned tiers are reviewed daily, last 30 days, and have separate earn and keep thresholds, so a quiet month can lapse a tier. And the historical cutoffs advance over time and differ for each data type, so a window that works today will silently exclude data later. The documentation also warns that its own endpoint-cost list is the current one in effect, which is the clearest possible signal that the others are not permanent.





































