On-chain Analytics Platforms

What this category is

Products that read the chains themselves rather than the venues on top of them. The raw material is public and free; what is sold is indexing, labelling and the judgement embedded in a metric's definition.

The boundary against market data runs through where the number comes from, not what it describes. A token's price on a DEX is market data even though the trade settled on-chain. A token's holder distribution is on-chain analytics even though it moves the price. Products that do both are listed under the job they are bought for, with the other as a secondary category — which is why the listing below carries four market-data APIs alongside the seventeen cards that call this category home.

Four products, one category

Comparing across these four is close to meaningless; comparing within one is exactly what a buyer needs. The job facet on each card says which one it is, and the four groups are unevenly sized in a way that tells you where the money is.

A SQL editor over indexed chain data — Dune, Allium, Footprint Analytics and Bitquery. You write the query. The product is the indexing, the schema and the execution engine, and the ceiling is your own SQL. Buy this when the question you have is not one somebody else has already asked, and expect to spend time on the schema before you get an answer. Note that only three of those four actually speak SQL — Bitquery is GraphQL, which is a different skill and a different query you cannot lift from a colleague's dashboard.

A labelled dashboard — Nansen, Arkham and Bubblemaps. The product is the label set: the work of deciding that this address is a particular exchange's hot wallet and that one is a fund. You do not write queries; you read somebody's attribution. Buy this when the question is "who", and understand that you are buying a judgement you cannot audit. Labels are proprietary, rarely dated, and wrong often enough to matter.

A curated metric set is the largest group by some way — eight of the seventeen, including Glassnode, CryptoQuant, Santiment, Artemis and Token Terminal. Somebody has already defined the metrics they think are worth watching and computes them consistently. You get comparability over time and no ability to ask anything outside the list. Buy this when you want the same number next quarter that you got this quarter.

Protocol and chain accounting is the fourth kind and the one usually missed — DefiLlama and Trading Strategy on the DeFi side, L2BEAT and growthepie on the rollup side. These answer questions about a protocol or a network rather than about an address or a price, and the last two are the only cards in the category published as open source, under grant funding rather than a subscription. That is not a discount — it is a different incentive, and it shows in what they measure.

Nothing here is as fresh as it looks

Twelve of the seventeen cards record a latency that is not real-time. Five claim real time — Arkham, Nansen, Bitquery, Santiment and its API sibling — and two say end-of-day outright, which are Artemis and Token Terminal, the two most focused on protocol revenue. The rest sit in between at "delayed".

This is not a flaw and it is rarely disclosed as a lag. Indexing a block, decoding contract calls, resolving labels and recomputing a metric are four sequential steps, and a platform that recomputes a daily series has no reason to do it more than daily. What it means for you is narrow and absolute: a decision that has to be made inside a block cannot be made from this category. Front-running a whale is not a product here. Understanding what a whale did is.

The dashboard and the API are two products at two prices

The catalogue has a natural experiment sitting in it. Santiment ships two cards because it is two products — the dashboard and SanAPI, the programmatic access to the same numbers — and the cheapest paid month on the API card is more than three times the cheapest paid month on the dashboard. That is the shape across this whole category, usually hidden inside one pricing page rather than split across two.

If you intend to pull numbers into your own system, price the API tier, not the one on the front page. The gap between the two is routinely larger than the gap between free and paid.

The pricing that is not published follows the four kinds above rather than the vendors. Every one of the query platforms keeps a tier with no number on it, and Allium publishes no number anywhere. None of the three labelled dashboards does — Nansen prints a per-call rate next to its monthly plans, Arkham prices its intel API by the month, and Bubblemaps charges nothing for the map at all. What Bubblemaps does instead is reserve its advanced features for holders of the vendor's own token, with no threshold published, which is a cost that never shows up in a price comparison because it is not a price; a token threshold is not a price is the long version. The per-category counts are on tools you cannot buy without a sales call, and why the silence clusters where it does is in why this market won't quote you a price.

The mistake this category invites

Two platforms' versions of the same metric can differ by more than the thing you are measuring. "Active addresses" can mean senders, receivers, or either. Exchange netflow depends entirely on whose label set says which addresses belong to the exchange, and those label sets are built by different teams from different evidence. TVL may or may not net out collateral counted twice across two protocols.

None of that is error. It is two methods, and the numbers are not interchangeable even when they carry the same name — the same problem the price side of the catalogue has with aggregate prices, one layer down. The practical rule is to pick one platform per series and stay on it, because a chart that switches sources mid-way is an artefact with a trend drawn through it. Glassnode versus CryptoQuant is the two-vendor version of this argument in full.

What to check before you commit

Which chains, at what depth. Every platform here claims multi-chain. Depth varies enormously: a chain may be indexed to full contract-call level, or it may be balances and transfers only. The chain you care about is either a first-class citizen or a line in a coverage table, and the table does not distinguish.

Entity labels go stale silently. An address labelled as an exchange wallet two years ago is still labelled that way after the exchange stopped using it. Ask when the label set was last revised, and treat an unlabelled address as unlabelled rather than as retail.

Historical recomputation. When a platform revises a metric's definition, some backfill the history and some do not. If you are charting a two-year series, that decision is the difference between a trend and an artefact.

What the credit actually buys on a query platform. A metered plan on a SQL product is charged against execution, not against calls, so an unselective query costs what a careless index costs. The meter, the window and the penalty are three separate rate-limit choices, and they are made independently.

Whether the output may leave your organisation. A research subscription and a redistribution licence are different purchases here exactly as they are for price data, and the clause is in the terms rather than on the pricing page.

What is not here

Block explorers, node providers and indexing infrastructure sold as a developer backend. Reading a single transaction is not analysis, and an RPC endpoint is plumbing rather than a market tool. Wallet and custody products are out of scope for the catalogue entirely. Anything whose output is which token to buy is advice, not tooling.

One more thing belongs in this section, because it is a fact about the category rather than about any card in it. Three on-chain platforms have pages here written for readers whose product has already gone, and all three shut down outright rather than changing their terms: Flipside, IntoTheBlock and Parsec. Read that as a warning about dependency rather than about any vendor on this page: a dashboard is a habit, but a query you scheduled, a metric you charted for two years and a label set you cited are all things that stop existing on somebody else's timetable.