ChainlakeAsk your on-chain data anything.
Chainlake's AI agent writes the SQL, builds the model, and shows its work. Starting with BSC, streaming in seconds -- with every query on the record, not a black box.
Private beta — built for funds, protocol teams, quants, and analysts working with on-chain activity.
Which BSC wallets moved into USDT the fastest after the last CEX outflow spike?
How it works
From a question to a trusted answer
Every step the agent takes is real, not simulated -- and every step is visible, not hidden behind a single summarized answer.
01
Ask
Ask about wallets, tokens, or protocol activity in plain English -- no SQL required to start.
02
Explore
The agent looks up schema, decides whether it needs a query, and runs read-only SQL against a whitelist -- with timeouts and scan limits, never DDL/DML.
03
Model
Recurring logic becomes a reusable metric -- an aggregation and expression the agent applies consistently, instead of re-deriving the same number a different way each time.
04
Visualize
Results become a chart, saved into a new or existing dashboard -- not a one-off image you lose after the session.
05
Trust the answer
You get a plain-language conclusion. Every query, parameter, and timing that produced it is on the record in Trace -- open it any time.
time_to_rotate_usdt
min(block_time) grouped by wallet, filtered to USDT transfers after the outflow spike
Why BSC first
Built and stress-tested on one of the fastest chains
BSC produces a block roughly every 0.45 seconds -- a demanding baseline for proving out data freshness. Our ingestion engine, chainlake-flow, is a self-built streaming runtime, not a one-off scraper wired to a single chain.
Seconds-fresh, streaming
Real-time streaming ingestion from BSC with fork-aware recovery and dead-letter retry -- current end-to-end freshness is on the order of seconds.
Built for more than one chain
chainlake-flow was refactored from an EVM-only runtime into a multi-chain adapter framework. More EVM chains are next on the roadmap; non-EVM chains follow after that.
One model, every chain
The goal is querying the same table or model across chains without rewriting anything per chain. See the roadmap below for where this stands.
Beyond on-chain data
Bring your own data in, too
On-chain analysis is the core of Chainlake, not the only thing it does. Connect your own warehouse or database and ask the same agent to reason across both.
ClickHouse
Connect today
PostgreSQL
Connect today
MySQL
Connect today
More connectors are on the way.
What's next
What we're building toward
Told plainly: these aren't shipped yet. We'd rather you know exactly where the line is today.
Sub-second freshness
The Kafka/Pinot streaming path is deployed in our infrastructure today. It is not yet exposed as a queryable source in the product -- current end-to-end freshness is seconds, not sub-second.
More EVM chains next, non-EVM after
Our ingestion engine (chainlake-flow) was refactored from an EVM-only runtime into a multi-chain adapter architecture. Additional EVM chains are next on the roadmap; non-EVM chains (an early, experimental Sui adapter already exists) follow after that.
One model, every chain
The goal: query the same table or model across chains without rewriting anything per-chain. This is the direction chainlake-flow is built toward, not a shipped capability yet.
Knowledge for the agent
A dedicated place to give the AI Agent durable context and reference material, beyond what it discovers per-session.
Trust
Built to hand real data to
The agent only ever runs read-only queries, scoped to the permissions of the person who asked -- the same access rules as manual querying, not a shortcut around them.
- Passkey and MFA-based sign-in, enforced per organization
- Org -> project permission model with scoped resource sharing
- Organization-level audit log alongside per-user activity history
- Multi-tenant by design -- one account, multiple isolated organizations
Private beta
Ask your first question about BSC
We're onboarding funds, protocol teams, and analysts directly while real-time chain data finishes rolling out. Request access and we'll follow up.