Predexon

What it is

Three products behind one key, and they are not equally deep. The tick archive is the newest and the one worth the card: every order-book placement, update, cancel and fill, as raw deltas in compressed Parquet, downloadable in bulk rather than paged over HTTP. The WebSocket runs eight channels on Polymarket — confirmed trades, pending trades from the Polygon mempool, splits and merges, lifecycle events, L2 order book, UMA oracle proposals and disputes, pUSD collateral movements, and Chainlink crypto price ticks. The REST API is a conventional normalized feed over five venues with wallet P&L, leaderboards and smart-money analytics indexed from the chain.

The venue spread is wider than the coverage. Markets, order-book history and trades exist for Polymarket, Kalshi, Limitless, Opinion and Predict.fun; the tick archive covers three of those; the WebSocket, the oracle feed, the wallet analytics and the builder-fee endpoints are Polymarket only. Read the endpoint directory before assuming a capability travels across venues.

Base URLs are api.predexon.com for REST and wss.predexon.com for streaming, with the key in an x-api-key header. There is also an x402 pay-per-call route through an outside facilitator, for agents and one-off pulls that would rather not hold an account.

Availability

No geographic gate and no identity check: an account is a sign-up at the dashboard, and the vendor also offers a pay-per-request path that needs no account at all. Nothing on the service is a venue, so there is no jurisdiction question about the data itself — only about whether you can trade on what you read.

The service is live and versioned. On 19 September 2026 a request to /v2/{venue}/markets for each of the five venues answered 401 missing_api_key, and a call to the old /v1/markets path answered 410 api_version_deprecated with a pointer to v2 — which is a better sign of a maintained API than a marketing page.

Pricing

Two meters that do not talk to each other.

The request plans are Free at 0 USD, Dev at 49, Pro at 249 and Enterprise from 499. What they buy is rate and premium-request quota: 1, 20 and 100 requests a second, with 1,000, 1,000,000 and 5,000,000 premium requests a month and overage at 5 and 2.50 USD per 100,000. The Free plan has no WebSocket at all and no smart-money or Binance endpoints. Pro carries a three-day trial.

The claim that carries the most weight is that historical data is free of quota on every plan: the vendor states that all list-market and order-book-history endpoints across every venue are unlimited and never accrue usage. Free of quota is not free of an account — every one of those paths returned 401 without a key when checked.

The tick downloads are billed against a separate prepaid data-credit balance, by the gigabyte of compressed Parquet you actually pull: 40 USD per GiB for Polymarket, 80 for Kalshi ticks, Kalshi book snapshots and Opinion. The vendor publishes its own worked sizes — a full day of Polymarket is about 13 GiB, or roughly 520 USD, and a day of Kalshi 3 to 5 GiB. Every download has a 5 USD minimum except Kalshi book snapshots, a download over 100 USD must carry a quote id obtained first, and market_id-filtered pulls are priced on the pruned files, which makes a single-market pull the cheap way to check the data before committing. Adding a payment method grants 50 USD of credits once per organization, top-ups run from 5 to 10,000 USD, and a daily spend cap can be set on the dashboard.

Markets & resolution

Whatever the five venues list, which in practice is elections, sports, macro releases and crypto prices. Predexon resolves nothing and settles nothing, so resolution_source here is null.

What it does do, uniquely in this catalogue, is expose resolution as a data series. The UMA endpoints list Polymarket questions filtered by oracle state and return an event timeline per market, and the oracle WebSocket channel streams proposals, disputes, settlements and resets as they happen. For anyone whose exposure is to the resolution rather than to the price — a disputed question is four to six days of locked capital — that is the feed that matters, and no venue API in this category publishes it in that shape.

The tick archive is finalized on a T-1 schedule: each UTC day is compacted the following morning around 06:00 UTC, a request for a day not yet compacted returns 404 not_final, and a single request spans at most 31 days.

Integrations

REST, WebSocket, bulk Parquet and MCP. The MCP server is predexon-mcp on npm, MIT-licensed, at 0.3.0 last published 1 August 2026, run with npx predexon-mcp and an API key in the environment; it exposes the Data API endpoints as typed tools. There is no first-party Python or TypeScript SDK — the documentation shows plain requests and fetch calls, which for a key-in-a-header API is honest rather than lacking.

Two documentation pages do work that most vendors here skip. A migration guide gives side-by-side translations from Polymarket's native API, Goldsky, Dune and self-built indexers to the equivalent Predexon endpoints, which is the fastest way to judge whether this replaces something you already run. And a best-practices page covers pagination, retries, rate-limit handling, WebSocket reconnection and idempotency rather than leaving them to the reader.

Limitations

• Nothing reads without a key, including the endpoints the vendor calls free. There is no demo endpoint and no delayed public tier, so the coverage claims on this card are the vendor's.
• The archive starts in March 2026. There is no order-book history before then at any price, and Kalshi book snapshots only begin on 18 September 2026 — the day before this card was written.
• Coverage is lopsided towards Polymarket. The WebSocket, the UMA oracle feed, the wallet and smart-money analytics and the builder endpoints are Polymarket-only; the other four venues get markets, books and trades.
• The tick archive is expensive at scale and priced by egress. A full day of Polymarket ticks is roughly 520 USD by the vendor's own example, so a year of it is not a subscription decision.
• No WebSocket on the free plan, which removes the product's headline feature from the tier a reader would evaluate with.
• A mempool observation is not a fill. The pending-trades channel is a leading indicator with a failure rate, and the documentation says so; anything built on it needs an explicit reconciliation path when a pending trade never confirms.
• The v1 API is already gone, retired behind a 410. That is good practice and also a reminder of how fast this surface moves.

Alternatives

DepthFeed is the other order-book archive and goes the other way — narrower scope, crypto up-or-down windows on three venues, no bulk files, but a keyless demo and a 29 USD entry plan. FinFeedAPI serves historical trades, quotes and raw limit-order events from flat files across nine venues. Predictefy covers sixteen venues with candles, books and cross-venue matching but sells no archive download. Marketlens sells Polymarket book history by the row rather than by the gigabyte, from 39 USD a month, with a backtester that replays it with queue priority. Polyrama and the Dune Polymarket dashboards compute wallet P&L from the chain for free, if that rather than the book history is what is wanted.