Pyth: what the oracle is and what you use it for

An oracle bridges off-chain facts into on-chain programs. Smart contracts cannot “call Binance”; they read accounts or messages someone put on the ledger. Pyth is one of the main oracle networks in crypto: a global price layer built around first-party publishers (exchanges, market makers, trading firms, and similar venues) that contribute market data, which applications then consume on Solana and 100+ other chains.

Docs hub: docs.pyth.network. Product surface today is broader than “SOL/USD”: crypto, equities, FX, and indexes appear in the feed catalog, with Core and Pro tiers and an upcoming Core upgrade path (API keys for Hermes from 18 Aug 2026 — check the upgrade guides if you integrate near that date).

Click to zoom

What Pyth is (mechanically)

Official framing: Core targets decentralized, multi-chain delivery with pull and push (docs cite ~400ms-class update frequency and broad chain coverage). Pyth Pro targets ultra-low-latency / customizable channels for venues that need tighter market-data product features. Feed IDs and contract addresses are listed in the developer hub.

Typical Solana pull flow
1. Client/crank: GET signed price update from Hermes (feed id)
2. Tx ix: post update to Pyth receiver / price account
3. Same tx (or immediately after): your program reads price + conf
4. Enforce: max confidence, max age, maybe ema vs spot rules

Safe use is not “read a number.” It is price + confidence + age + stall rules. Wide confidence or stale publish time should fail closed on risk engines.

What people use Pyth for today

On Solana specifically, pull integrations are the usual path for protocols that need the price in the transaction that moves risk — liquidations, opens, collateral adjusts — so nobody relies on a stale account someone else paid to update hours ago.

What else it could be used for

Same primitive, wider product surface — if the feed catalog and confidence behavior fit the risk:

• On-chain funds and tokenized treasuries — NAV marks, fee accrual, gated subscriptions at oracle time
• Prediction / insurance settlement — resolve on published index/FX/equity feeds where governance accepts Pyth as source of truth
• Dynamic fees and bonding curves — fee tiers or mint prices that track an external reference
• Cross-margin and portfolio margin — correlated assets under one oracle family with explicit confidence haircuts
• Agent / intent systems — agents that only sign if on-chain mark is inside a band (combine with cranks / TukTuk for scheduled checks)
• Compliance-shaped products — circuit breakers when confidence blows out or publish stalls

Adjacent product: Pyth Entropy — verifiable RNG aimed at EVM games, mints, lotteries (commit-reveal). Useful when your randomness need is EVM-side; Solana apps often use other VRF/RNG stacks (including Switchboard randomness). Don’t conflate Entropy with price feeds.

Design rules when you build on Pyth

• Check confidence — treat wide bands as “don’t liquidate / don’t mint”
• Check publish time / slot age — reject stale updates
• Prefer pull in the same tx for risk-critical paths
• Know feed IDs per environment; don’t hardcode the wrong cluster’s account
• Plan for Hermes / API key policy changes (Core upgrade timeline)
• Separate trading mark from conservative collateral mark if your risk policy needs haircuts beyond conf

Alternatives

No oracle is universal. Choose on feed coverage, update model, customizability, and failure modes.

Practical Solana split many teams use: Pyth for majors and liquid cross-venue assets; Switchboard when you need a custom job or randomness; pool TWAP only with strict caps and manipulation-aware risk; never a single thin pool spot for large collateral.

Quick chooser

Related on DevRels

• Switchboard: customisable oracle
• Cranks and TukTuk (scheduled oracle posts)
• Perps patterns

Summary

Pyth is a first-party market-data oracle network: publishers contribute prices, apps pull (or push) verified updates on-chain, and programs act on price plus confidence plus freshness. Builders use it for marks, liquidations, collateral, synthetics, and multi-asset references across many chains. It is not the only oracle — Switchboard covers custom data, other networks dominate some ecosystems, and pool TWAPs remain a last resort for assets with no honest external market. Match the feed to the risk, and fail closed when confidence or time says the number is not safe to use.