Sound of AVM ships v0.9.0, sonifying Algorand's live mempool in the browser
A Network You Can Hear
Algorand settles a new block every few seconds, each one sealing whatever the network just did — a stablecoin payment, an asset transfer, a smart-contract call. Most people watch that traffic as scrolling tables of hashes on a block explorer. Sound of AVM listens instead: a browser synthesizer that turns live on-chain activity into music, so a busy network sounds like a busy room and a quiet one like silence with an occasional note. Its own description is the most precise one available: 'Nothing is pre-recorded; if you hear a note, a transaction just happened.'
The app reads the mempool — the pool of unconfirmed transactions waiting for the next block — rather than block history, so it hears transactions as they propagate, before settlement. Each instrument is a synth card pairing a rule, which transactions it cares about, with a sound to play when one arrives. On July 26, 2026 the project shipped v0.9.0, its first numbered release; its in-app guide offers the fastest onboarding in the tool: load a preset, press Start, and the spacebar stops and starts the music from anywhere on the page.
One Synth per Transaction Type
A synth's first choice is the kind of transaction it listens for. Algorand's full transaction set is on the menu — pay, axfer, appl, acfg, keyreg, afrz, stpf, hb, group, block — and each can be narrowed by subtypes. A pay synth can filter on amount, sender or receiver; an axfer synth adds asset ID, opt-in, opt-out and [clawback](/glossary/clawback "A feature in some blockchain assets allowing an authorized address to forcibly reclaim tokens from a user’s wallet. It introduces centralized control "); an appl synth can watch a specific application. The filters are where the tool gets surgical:
| Concept | Real-World Implication |
|---|---|
| pay — plain ALGO payments | An amount range turns the voice into a whale detector: a deep note that only sounds when a payment exceeds, say, 5,000 ALGO. |
| axfer — asset transfers | Asset boxes search by name as well as ID, so typing 'USDC' is enough to make a voice play only when that token moves. |
| appl — smart-contract calls | Track a single application, or the foreign assets and accounts its calls touch, without reading transaction logs. |
| group — grouped transactions | Pair a synth with group to catch a token being swapped on a [DEX](/glossary/decentralized-exchange "A peer-to-peer marketplace built on Algorand where users can trade ASAs directly from their wallets without relying on a centralized intermediary (lik"), since a swap is an atomic group, not a single transaction. |
| block — one note per block | The rhythm section: a sound fired once per certified round, giving everything else a steady beat. |
The rest is sound design. Five Tone.js engines sit under the voices — a single-voice synth for tight, percussive hits; a mono synth with its own filter for basslines; a polyphonic synth for overlapping pads; and AM and FM variants for metallic or bell-like timbres. An eight-step sequencer advances on transactions rather than a clock, so the riff's tempo is the network's: busy chain, fast riff. The guide frames pitch as composition — 'Put the busiest transaction type low and the rarest one high, and you get bass and melody for free.' When one synth is being triggered faster than its cap, Mode decides the outcome: Single drops the overflow, while Aggr collapses the burst into one longer, louder hit.
From Polling to a Live Feed
The project began at the Algorand Dev Retreat at Home as a prototype in plain HTML and JavaScript, born from the observation that a chain's transaction stream is a kind of rhythm. That first version polled pending transactions every 50 milliseconds through a public AlgoNode API or a personal node, fed the results to Tone.js, and asked users to be nice to the free API. Connecting a personal node meant flipping a CORS header in the node's config. It worked, with caveats — scheduling accuracy was hostage to the poll interval, and the API's rate limits made it a toy rather than a monitor.
v0.9.0 replaces the whole data layer. Instead of polling, the app opens a WebSocket to Nodely's mainnet gossip gateway and decodes signed transactions in the browser as relays broadcast them. The code is explicit about the trade: 'Every transaction arrives as it propagates, with its real arrival time — no polling, no count-diffing, no tokens.'
| Concept | Real-World Implication |
|---|---|
| Prototype: 50 ms polling of a public API | Notes arrive quantized to the poll interval, and the project estimated free-tier rate limits capped continuous listening at roughly five hours. |
| v0.9.0: keyless Nodely gossip WebSocket | A note fires as the transaction propagates to relays, with no API key and no local node required. |
| Atomic-group buffering (~300 ms) | The members of a grouped transaction are collected and flushed as one event, so a swap sounds like a single gesture rather than a stutter. |
That last detail matters for listeners as much as the synthesis: the client implements a slice of Algorand's relay wire protocol in the browser, subscribing to the transaction message tag, decoding its msgpack payloads, and remembering recently seen transactions across two generations so relay rebroadcasts don't double-trigger a note.
Presets You Can Mint On-Chain
Layouts are portable. A preset is a JSON file describing every synth and setting, saved locally or exported to share; v0.9.0 also encodes presets and minted preset tokens into the address bar, so a URL is enough to pass a sound along. The Wallet button exists for one reason: minting a preset on-chain as an 'NFPreset'.
The mechanism deserves unpacking because it uses Algorand's native token machinery directly. Algorand issues assets — Algorand Standard Assets, or ASAs — without needing a smart contract, and [ARC-3](/glossary/arc-3 "A technical standard for defining non-fungible tokens on the Algorand blockchain, specifying how metadata like the asset name, image, and audio links ") is the convention for NFTs among them: the asset's URL points to JSON metadata hosted on [IPFS](/glossary/ipfs "A decentralized file storage system where data is stored across multiple computers, making it resistant to censorship or loss. NFTs often use IPFS to "), marked with a '#arc3' suffix. An NFPreset is an [ASA](/glossary/algorand-standard-asset "A built-in mechanism that allows anyone to create and issue new tokens (like stablecoins, utility tokens, or NFTs) directly on Algorand's base layer, ") whose IPFS metadata embeds the preset as a compressed base64 string in its properties. Loading one means pulling the asset's metadata from the indexer, fetching the file from IPFS, and decoding it back into voices. The roadmap describes the mint as a single wallet-signed asset-creation transaction, with a marketplace for trading NFPresets deferred to a future v2.
The format has been exercised on-chain. The asset 'block anxiety' (unit name NFP, total supply 1) was created on August 5, 2025 from an address funded by hampelman.algo, the developer's on-chain handle — matching the example metadata in the project's own roadmap document — and was followed by two further mints of the same preset, labeled NFP2 and NFP23. The conceit is quietly strange and rather functional: an NFT here is not an image but a playable instrument configuration, and trading one is trading sound design.
Small, Open, and Still in Tune
The numbered release also brought a row of visualizations that render the same data as pictures — Territory, in which every transaction claims its share of the screen and 'squeezes the standing ground aside,' plus Kintsugi Ledger and Music Box, alongside the earlier Score and Loom. The Record button, intended to capture output as audio, is not wired up yet.
The project stays honest about its own limits. The original README notes that 'Transactions come in too fast for Tone.js, making it unable to create unique timestamps for every one of them' — the reason aggregated mode exists — and there is no formal audit of the codebase, a typical state for an open-source tool of this scale. Its reach is modest: eight stars on the original repository in the Algorand Developer Retreat organization, one star on the actively maintained copy under the developer's personal account, and a single short write-up from an Algorand news aggregator in late July 2026 that described the older polling architecture — a sign the revamp has yet to find its audience.
It is the kind of project that only exists because a hackathon let one developer chase a strange idea: that a blockchain's activity has a tempo worth hearing. The code remains open and the maintainer has welcomed suggestions since the prototype days. For anyone who has found a block explorer hypnotic in the wrong way, listening to the network — bass for whales, melody for rare events, a block synth keeping the beat — is a genuinely new way to watch a chain work.
Source
- Sound of AVM
- The Sound of AVM — Algorand Developer Retreat organization (prototype)
- The Sound of AVM — active development repository
- PXke Algorand coverage
- Verifiably Random: S2:E8 - Inside the open-source coding at the Algorand Dev Retreat
- Post by @algorand.pxke.me — Bluesky
- RelayUrl path and query issues · Issue #735 · mikedilger/gossip
- QuickStart | Nodely
Source: editorial://brief/28e66d55-cc55-47c9-b153-9067eea0a838