Skip to content

locallaunchsc-cloud/solana-dev-skills / pyth-oracle

solana-pyth-oracle

Verified

Use this skill when the user wants to integrate Pyth price feeds into a Solana or Anchor program — find feed IDs, update on-chain price accounts (pull oracle model), read prices in instructions, handle staleness and confidence intervals safely.

Pyth Oracle

Overview

Pyth is the dominant oracle on Solana — hundreds of feeds (crypto, FX, equities, commodities) sourced from first-party publishers (Jane Street, Jump, CBOE, Wintermute, …) and aggregated on Pythnet. As of May 2026 the canonical integration on Solana is the pull model: prices live off-chain on Pythnet, you fetch the latest signed update from the Hermes web service, and your transaction posts it to the Pyth Solana Receiver program on the same Solana transaction that consumes it. The older push model (fixed Crypto.SOL/USD account at a known address) is deprecated for new integrations — the on-chain Solana sponsored price feed accounts have been progressively turned down since late 2024 and most tutorials online still show the dead path. This skill covers the pull model only.

When to use this skill

  • The user asks "how do I get the price of SOL/BTC/ETH/USDC in my Solana program"
  • They mention Pyth, oracle, PriceUpdateV2, pyth-solana-receiver-sdk, Hermes
  • They're building anything that touches USD value on-chain: lending, perps, AMM TWAP guard, liquidations, dollar-denominated mint price, collateralization check, payout in SOL of a fiat-denominated amount
  • They have stale-tutorial code calling load_price_feed_from_account_info against a hard-coded Crypto.SOL/USD pubkey and getting AccountNotFound

Skip this skill if the user wants sub-100ms latency for an HFT-style use case — that's Pyth Lazer, a separate product (@pythnetwork/pyth-lazer-solana-sdk), not covered here.

Prerequisites

  • Anchor 0.31.x or 0.32.x (the SDK officially supports 0.28, 0.29, 0.30.1, 0.31.1; current 1.x line tracks Anchor 0.32.x)
  • Rust crate (in your program's Cargo.toml):
    [dependencies]
    pyth-solana-receiver-sdk = "1.2.0"   # latest, tracks anchor-lang 0.32.x
    anchor-lang = "0.32.1"
    
    Note: if your program is still on Anchor 0.31.x, pin pyth-solana-receiver-sdk = "0.6.1" instead — that's the last release before the 1.x line bumped anchor-lang to 0.32.x. The import paths and structs used in this skill (PriceUpdateV2, get_feed_id_from_hex, get_price_no_older_than) are identical across 0.6.x and 1.x.
  • TS client (package.json):
    {
      "dependencies": {
        "@pythnetwork/pyth-solana-receiver": "^0.15.0",
        "@pythnetwork/hermes-client": "^2.0.0",
        "@coral-xyz/anchor": "^0.31.1",
        "@solana/web3.js": "^1.98.4"
      }
    }
    
  • A Solana RPC endpoint that supports sendTransaction with maxSupportedTransactionVersion: 0 (versioned tx are required — the post-update tx is large and uses ALTs)
  • Funded keypair on the target cluster (devnet/mainnet). A pull update costs ~0.001 SOL of transient rent that you reclaim by closing the price update account after consumption.

Workflow

1. Find the feed ID

Every feed is identified by a 32-byte hex ID, the same across every chain (Solana, EVM, Aptos, …). Look it up at pyth.network/developers/price-feed-ids or hit Hermes directly:

curl 'https://hermes.pyth.network/v2/price_feeds?query=SOL/USD&asset_type=crypto'

Top crypto feed IDs (verified May 2026):

PairFeed ID (hex, 0x-prefixed)
SOL/USD0xef0d8b6fda2ceba41da15d4095d1da392a0d2f8ed0c6c7bc0f4cfac8c280b56d
BTC/USD0xe62df6c8b4a85fe1a67db44dc12de5db330f7ac66b72dc658afedf0f4a415b43
ETH/USD0xff61491a931112ddf1bd8147cd1b641375f79f5825126d665480874634fd0ace
USDC/USD0xeaa020c61cc479712813461ce153894a96a6c00b21ed0cfc2798d1f9a9e9c94a

2. Fetch the latest signed update from Hermes

Hermes is the public web service that streams Pyth updates off Pythnet. It returns a base64 VAA (Wormhole-signed price message) that the on-chain receiver will verify.

import { HermesClient } from "@pythnetwork/hermes-client";

const hermes = new HermesClient("https://hermes.pyth.network/", {});
const SOL_USD = "0xef0d8b6fda2ceba41da15d4095d1da392a0d2f8ed0c6c7bc0f4cfac8c280b56d";

const { binary: { data: priceUpdateData } } =
  await hermes.getLatestPriceUpdates([SOL_USD], { encoding: "base64" });

priceUpdateData is string[] (one entry per VAA). Hold onto it for step 3.

3. Post the update on-chain

PythSolanaReceiver from the TS SDK builds a 2-stage transaction set: (a) post the update — this creates an ephemeral PriceUpdateV2 account owned by the receiver program, deterministically derived from the random keypair the SDK injects; (b) your consumer instruction reads that account; (c) optionally close the account to reclaim rent.

import { PythSolanaReceiver } from "@pythnetwork/pyth-solana-receiver";

const receiver = new PythSolanaReceiver({ connection, wallet });
const txBuilder = receiver.newTransactionBuilder({ closeUpdateAccounts: true });

await txBuilder.addPostPriceUpdates(priceUpdateData);

await txBuilder.addPriceConsumerInstructions(async (getPriceUpdateAccount) => {
  const priceUpdate = getPriceUpdateAccount(SOL_USD); // <-- pass to your program
  return [
    await myProgram.methods.consumePrice()
      .accounts({ priceUpdate })
      .instruction(),
  ];
});

await receiver.provider.sendAll(
  await txBuilder.buildVersionedTransactions({ computeUnitPriceMicroLamports: 50_000 }),
  { skipPreflight: true },
);

The same builder handles posting multiple feeds in one tx — pass an array of VAAs to addPostPriceUpdates and call getPriceUpdateAccount(feedId) for each one inside the consumer callback.

4. Consume the update inside Anchor

On the Rust side, declare a PriceUpdateV2 account in your #[derive(Accounts)] and call get_price_no_older_than. This single call does three things at once: deserializes, checks the feed ID matches what you expect, and rejects anything older than maximum_age seconds.

use pyth_solana_receiver_sdk::price_update::{get_feed_id_from_hex, PriceUpdateV2};

#[derive(Accounts)]
pub struct ConsumePrice<'info> {
    pub price_update: Account<'info, PriceUpdateV2>,
    pub signer: Signer<'info>,
}

pub fn consume_price(ctx: Context<ConsumePrice>) -> Result<()> {
    const SOL_USD_FEED: &str =
        "0xef0d8b6fda2ceba41da15d4095d1da392a0d2f8ed0c6c7bc0f4cfac8c280b56d";
    const MAX_AGE_SEC: u64 = 60;

    let feed_id = get_feed_id_from_hex(SOL_USD_FEED)?;
    let price = ctx.accounts.price_update.get_price_no_older_than(
        &Clock::get()?,
        MAX_AGE_SEC,
        &feed_id,
    )?;

    // price.price : i64   (scaled by 10^price.exponent, which is negative)
    // price.conf  : u64   (same scaling)
    // price.exponent : i32
    msg!("Price: {} ± {} × 10^{}", price.price, price.conf, price.exponent);
    Ok(())
}

5. Apply safety checks the SDK doesn't do for you

get_price_no_older_than enforces age + feed ID, but does not bound confidence. In production add an explicit confidence guard before using the value:

// Reject if conf > 1% of price — typical liquidation-engine threshold.
let conf_bps = (price.conf as u128)
    .checked_mul(10_000).unwrap()
    .checked_div(price.price.unsigned_abs() as u128).unwrap();
require!(conf_bps <= 100, MyError::PriceTooUncertain);

6. Convert the integer price to a real number

Pyth prices are i64 scaled by 10^exponent. exponent is negative (typically -8 for crypto). To express "this much asset is worth $X":

// value_usd = amount × price × 10^expo
//   where amount is in the asset's smallest unit (lamports for SOL).
// For SOL exponent=-8: 1 SOL × 150_00000000 × 10^-8 = $150.
let amount_lamports: u64 = 1_000_000_000; // 1 SOL
let scaled = (amount_lamports as i128)
    .checked_mul(price.price as i128).unwrap();
// `scaled` now has 9 (lamports) + 8 (Pyth) = 17 decimal places of precision.
// Divide by 10^17 to get whole USD; keep more precision if you store fixed-point.

Never use floats on-chain. If your downstream math needs sub-cent precision, keep the scaled i128 and let the consumer divide.

7. Close the price update account (optional but recommended)

The TS builder's closeUpdateAccounts: true flag appends a close instruction so the ~0.001 SOL of rent flows back to the signer. Leave it on unless you have a reason to keep the account around for the same slot.

Common pitfalls

  • Push-model rot. Hundreds of tutorials and Stack Overflow answers still tell you to use pyth_sdk_solana::load_price_feed_from_account_info against a fixed pubkey like H6ARHf6YXhGYeQfUzQNGk6rDNnLBQKrenN712K4AQJEG (legacy SOL/USD). These accounts are no longer being updated on mainnet for new feeds and the path is deprecated. If you see load_price_feed_from_account_info anywhere in 2026 code, replace it.
  • Forgetting to pass the price update account. With the pull model the account address is different every transaction (it's derived from a random keypair the SDK generates). You cannot hard-code it — your program must accept it as an Account<PriceUpdateV2> and the TS client must pass it in via the getPriceUpdateAccount(feedId) callback.
  • Trusting the account without checking the feed ID. Account<PriceUpdateV2> only checks that the account is owned by the receiver program. An attacker could substitute a PriceUpdateV2 for, say, DOGE/USD and your "SOL price" code would happily read it. Always pass the expected feed ID to get_price_no_older_than, which checks it inside the account data.
  • Stale price. Clock::get()?.unix_timestamp is the on-chain time. price.publish_time is when Pythnet observed it. A reasonable MAX_AGE_SEC is 30–60s; for liquidations or anything adversarial, go tighter (10–20s) — but then you must reliably get Hermes data into the same tx.
  • Ignoring confidence interval. During flash crashes or thin-market sessions conf widens to several percent of price. A naïve consumer that reads price.price and ignores price.conf will mark a position based on a spread mid that no liquidity actually exists at. Always enforce conf / price ≤ threshold (1% is a common bar).
  • Exponent direction. exponent is negative (-8 for crypto). New devs sometimes multiply by 10^expo instead of dividing, which makes a 150 SOL look like 1.5 × 10^-6 instead of $150. The formula is real_value = price * 10^exponent.
  • Hermes rate limits. The public hermes.pyth.network is shared and rate-limited (no published limit, but production workloads will hit 429s at sustained traffic). For real volume run your own Hermes (docker run pythnetwork/hermes) or use a managed provider.
  • Pull update transaction size. Posting a VAA + your consumer instruction often exceeds 1232 bytes — that's why the SDK returns VersionedTransaction[] (sometimes multiple), uses an address lookup table, and recommends skipPreflight: true. Don't try to flatten it into a legacy Transaction.
  • EMA vs spot. price_message also carries ema_price / ema_conf (exponentially-weighted moving average). For mark-to-market on derivatives, the EMA price is the safer field — it's smoother and harder to manipulate within a single slot. Spot is what get_price_no_older_than returns by default.

References

Example

See scripts/anchor-program.rs for the on-chain Anchor consumer (deserialize → verify feed ID → check staleness → bound confidence → log).

See scripts/update-and-consume.ts for the matching TS client (fetch from Hermes → post via PythSolanaReceiver → invoke your program in the same tx).

reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 470kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall