CoinStats CEO Narek Gevorgyan on Building a Crypto AI Agent for Real-Time Research

Crypto analysis has change into tougher because the market has grown throughout chains, protocols, wallets, exchanges, and social platforms. A single funding resolution can require hours of checking token knowledge, on-chain flows, sentiment, information, liquidity, and portfolio publicity.

CoinStats began as a portfolio tracker, giving customers one place to watch belongings throughout wallets and exchanges. The firm is now constructing a extra bold product round crypto-specific AI, developer APIs, and agent-ready knowledge entry.

In an unique interview with BeInCrypto, Narek Gevorgyan, Founder and CEO of CoinStats crypto tracker, mentioned why crypto wants domain-specific AI, how CoinStats AI approaches analysis, and why machine-readable crypto knowledge will change into important as AI brokers enter the market.

CoinStats started as a portfolio tracker. What led the push towards an AI-driven crypto analysis product?

Tracking a portfolio is the straightforward half. The laborious half is knowing what to do subsequent.

Our customers have been spending hours leaping between X, Discord, Etherscan, information websites, analytics dashboards, and alternate pages simply to make one resolution. CoinStats already had the info layer in place, together with protection throughout 120+ chains, market knowledge, on-chain flows, and social context.

AI was the pure subsequent step. Instead of giving customers extra dashboards, we wished to assist them attain higher solutions quicker.

You are making a robust case for domain-specific AI in crypto. Where do general-purpose fashions nonetheless fall brief for critical crypto analysis?

It largely comes all the way down to structure and knowledge.

When a person asks CoinStats AI a query, specialised sub-agents work in parallel. One can pull real-time information. Another can scan social sentiment. Another can learn on-chain knowledge throughout 120+ blockchains. Another can verify alternate metrics. Another can analyze the person’s precise portfolio.

Those brokers report again, and the system synthesizes the data into one reply with interactive tables and charts, as a substitute of a lengthy wall of textual content. The mannequin is studying from stay sources moderately than recalling info from coaching knowledge. That reduces a giant a part of the hallucination threat.

We additionally let customers select the depth of the reply. CoinStats AI has three modes. Deep Research is for full multi-source reviews. Backtesting helps customers check methods in opposition to historic knowledge. Fast Mode is for fast lookups.

A basic mannequin often offers one fashion of reply. Crypto analysis has many various query sorts.

We tune CoinStats AI across the precise work crypto customers do, together with token analysis, pockets evaluation, threat checks, good cash monitoring, whale exercise, contract deployments, KOL sentiment, and macro correlations between issues like Fed coverage and ETF flows. The distinction turns into apparent as soon as the questions change into particular.

CoinStats has recommended its AI performs strongly in opposition to bigger basic fashions on crypto analysis duties. What precisely is it doing in a different way underneath the hood?

It is a mixture of stay knowledge entry, retrieval, task-specific brokers, and crypto-native reasoning.

General fashions often lack stay on-chain knowledge, so they can not reliably let you know who’s accumulating a token or the place liquidity is transferring. They additionally lack real-time market and social context, to allow them to miss narratives as they type. Their coaching knowledge can change into stale in a short time in a market the place a token can launch and transfer aggressively inside days.

They additionally cause like generalists. Crypto analysis usually requires understanding MEV, slippage, bridge threat, liquidity fragmentation throughout chains, pockets habits, alternate flows, and protocol-specific threat.

Privacy is one other main level. When a person pastes a pockets tackle into a basic AI mannequin, they could be exposing their holdings to a third-party supplier. Crypto customers care about this.

That is why we constructed Private Mode in CoinStats AI. When customers flip it on, queries are routed by Venice AI’s encrypted, decentralized system. No third-party AI supplier sees the person’s knowledge. Whether somebody is researching wallets, analyzing token flows, or wanting into positions they like to maintain non-public, the data stays between the person and the blockchain.

General fashions are helpful for informal questions. Serious crypto analysis wants stay knowledge, privateness, and crypto-specific context.

How are you desirous about accuracy, belief, and hallucination threat when customers could act on the output?

Crypto is a market the place unfastened accuracy can change into costly.

Our method is constructed round three ideas. First, each declare ought to be grounded in stay knowledge with sources, so customers can verify the work. Second, Backtesting Mode lets customers validate a thesis in opposition to historic knowledge earlier than risking capital. Third, we’re very clear concerning the product’s function.

CoinStats AI is a analysis software. It is constructed to assist the DYOR course of, not change person judgment. DYOR ought to be a part of the product expertise itself, not a disclaimer on the backside of a web page.

On the developer aspect, CoinStats can also be pushing its API and MCP assist for AI brokers and IDEs. Why is a developer-accessible crypto knowledge layer vital?

Crypto has a structural knowledge drawback. The info wanted to know a portfolio, market, or on-chain occasion is fragmented throughout lots of of chains, 1000’s of protocols, dozens of centralized exchanges, and a rising DeFi ecosystem.

Any developer or AI agent attempting to cause about crypto has two choices. They can spend years fixing aggregation themselves, or they’ll plug into a supplier that already does it.

That is the function we see CoinStats enjoying. We have spent years normalizing knowledge throughout 300+ exchanges and wallets, each main chain, and a lengthy tail of DeFi positions.

By exposing this by the CoinStats Crypto API and MCP server, builders constructing AI brokers, buying and selling instruments, analysis merchandise, or aspect tasks in environments like Cursor or Claude Code can entry portfolio state, market knowledge, information, and on-chain context as usable primitives.

They don’t must rebuild the pipeline earlier than constructing the product.

MCP is changing into a critical dialog in AI tooling. How do you see CoinStats becoming into a future the place crypto workflows are more and more dealt with by brokers?

Crypto knowledge has all the time been fragmented. Prices stay on one platform. Wallet balances stay elsewhere. DeFi positions sit throughout many protocols. NFTs could sit elsewhere once more.

For builders, stitching all of this collectively is usually the toughest half. Teams can spend extra time on knowledge plumbing than on the person expertise.

That is what we got down to remedy with CoinStats API. The protection spans 100,000+ cash and 200+ exchanges. It extends throughout 120+ blockchains. DeFi positions are resolved throughout 10,000+ protocols on the pockets stage.

Developers get one entry level for the total image. That adjustments what a small workforce can construct shortly.

Our MCP server takes the identical concept additional. AI brokers and LLMs can question pockets, DeFi, and portfolio knowledge straight. An IDE-integrated agent can pull a person’s positions, analyze them, and assist a workflow with out customized adapters.

This is vital as a result of crypto instruments are evolving. Future workflows will contain brokers monitoring threat, rebalancing portfolios, surfacing alternatives, and supporting analysis. For that to work, the info layer must be machine-readable, dependable, and full sufficient to know what somebody owns and the way these belongings transfer on-chain.

Pricing knowledge alone just isn’t sufficient. Agents want pockets knowledge, DeFi place decision, and long-term historic context. That is the layer CoinStats API is constructing.

Crypto development additionally relies upon on how simple it’s to construct helpful merchandise. Every hour builders save on knowledge aggregation could be spent bettering the person expertise.

If now we have this dialog once more a 12 months from now, what would you wish to have constructed, improved, or confirmed about CoinStats by then?

There are three issues I might wish to see.

First, I might need CoinStats to widen its lead in crypto-specific analysis in opposition to general-purpose AI. The benchmark we launched this 12 months is the start. We wish to increase it, run it extra continuously, and maintain the methodology open supply so anybody within the business can reproduce it.

The aim is to not win a single benchmark. The aim is to show that vertical AI constructed for crypto performs higher over time as a result of it has the correct knowledge, instruments, and reasoning atmosphere.

Second, I might need CoinStats API to change into the default crypto knowledge and analysis layer for the agent ecosystem. Between MCP, our x402-powered API, and our portfolio intelligence system, any agent that wants crypto context ought to have the ability to plug into CoinStats.

Third, I might need CoinStats to go farther from analysis into motion. Understanding why one thing is transferring is half the job. Helping customers act on these insights safely, inside the identical workflow, is the subsequent product frontier.

The finish aim has stayed the identical. Every crypto holder ought to have a private workforce of analysts working for them 24/7. A 12 months from now, I would like CoinStats to be a lot nearer to delivering that have by CoinStats AI.

The publish CoinStats CEO Narek Gevorgyan on Building a Crypto AI Agent for Real-Time Research appeared first on BeInCrypto.