Open-Source Intelligence · Crypto · 2026
Every wallet, every account, every launch leaves a trail.
CARLI is an OSINT agent. It reads public records the way a field analyst would — on-chain transactions, domain registrations, account histories, social signals — then lets Claude connect what they mean.
No logins. No subscriptions. Public data only.
Try it — click an example, no signup
First, the basics
Open-Source Intelligence is the practice of drawing conclusions from public records — data anyone can access, but few know how to read together.
A field analyst doesn't hack anyone. They pull the registration date here, the account age there, the funding trail underneath, and read the pattern the pieces form. CARLI does exactly that for crypto — at machine speed, with Claude reading intent.
No logins. No private data. No breach corpus. Only what is already public.
Public signals → one verdict
On-chain transactions
wallets, transfers, contracts
Domain registrations
WHOIS, SSL certificate history
Account histories
age, username changes, growth
Social signals
bios, links, posting behavior
level · confidence · signals
The premise
Nobody scams in a vacuum. They leave a wallet, a handle, a pattern.
Bubble maps see the wallets. Screeners see the volume. Neither reads the account behind the launch or the story being sold. That gap is where the money disappears. CARLI works in that gap.
What CARLI is for
Each of these is a live pattern on X and pump.fun right now. Each one is something CARLI is built to catch.
Viral · Mar 2026
Networks of bought X accounts post fabricated breaking news to farm reach, then pivot straight into pump-and-dumps. One account in a recent network had changed its username sixteen times in two years. That history is public. CARLI reads it in seconds, not the days it takes a human investigator.
Ongoing threat
A verified account with hundreds of thousands of followers gets phished, then drops a memecoin minutes later. By the time anyone notices, the token has already pumped and dumped. The tell is simple: did this account ever talk about crypto before today? CARLI checks that instantly, before you ape.
Invisible to every tool
Modern rug pulls run on narrative. A dying dev, fees going to family, a cause you can't say no to, then the wallet empties. No bubble map or screener can read whether a project's story follows the manipulation playbook. Reading intent from language is exactly what an AI agent can do, and tooling cannot.
The scale problem
Tens of thousands of tokens launch daily and the overwhelming majority end badly. The on-chain side has tooling. The social side, account age, coordinated shilling, posting behavior, crypto history, is still checked by hand, one tab at a time. CARLI is the layer that has been missing.
The instruments
Four instruments, one directive: tell you what something really is before your money is in it. Real data, read by Claude, returned as a verdict you can act on.
@handleAccount age against follower growth, engagement authenticity, posting behavior, and whether this account has any crypto history at all. Classifies account type first, then verdict: official, unverified, mismatch, or impersonation.
Solana · EVM addressBehavior pulled from public chain data via Helius and Etherscan. Dev wallet, whale, or flipper. Funding source, transaction history, and connected projects. An interpretation layer on top of the explorer, not another raw feed.
name · contract · domainDomain age and registration, linked social accounts, team identity signals, and connected wallets, scored end to end: safe, DYOR, high risk, or likely rug. Narrative analysis detects manipulation patterns. Built to be shared as a card.
free-form chatTalk to CARLI directly. "Check this account before I buy." It runs the full investigation conversationally and synthesizes one answer instead of handing you ten dashboards.
Under the hood
CARLI is an agent, not a wrapper. Each investigation is a chain of real tool calls that Claude orchestrates and then reasons over. The instrument does not describe the path. It walks it.
detect_input_type()You paste a handle, an address, a contract, or just a sentence. CARLI classifies the input and decides which instruments to fire. No forms, no choosing the right tool yourself.
claude.tool_use → [ chain, social, domain ]Claude chains real API calls in a single turn. On-chain reads from Solscan and Etherscan, social signals from the account graph, domain and registration from crt.sh and WHOIS. Live data, not cached guesses.
correlate(signals)The wallet that funded the deployer, the account that shilled it, the domain registered the same week. Individually these are noise. Linked together they are a pattern, and the pattern is the finding.
extended_thinking()For the hard part, the narrative, Claude reads the actual language of the project and the account against known manipulation playbooks. This is the step no screener can do, because it is about meaning, not metrics.
verdict := { level, confidence, evidence }Every finding gets a confidence grade, tentative, firm, or confirmed, with the evidence attached. The output is one structured verdict, scannable in seconds, not a wall of raw data to interpret yourself.
Public data only
CARLI reads what is already open: on-chain records, public accounts, registration data. No breach corpus, no private scraping. It is a due-diligence instrument, not a surveillance one.
The token
CARLI stays free, rate-limited per user. $CARLI aligns the people who benefit from the tool with the tool itself. Every rug that goes viral is a reason CARLI exists.
Run your DYOR
Paste an account, a wallet, or a project. Get a verdict in seconds. The trail is already there. CARLI just reads it.
No logins. No subscriptions. Public data only.