Inside Nansen's AI Trading Agent Platform

Nansen began as an on-chain analytics company and has labeled more than 500 million blockchain addresses, which it describes as its core competitive advantage. The labeling system combines agentic AI teams, algorithmic heuristics, and internal human attribution specialists, with every label requiring compiled evidence before entering the database. Nansen runs internal attribution evals using an LLM-as-a-judge approach where secondary agents evaluate primary agents and humans then evaluate those judging agents. Deterministic smart contracts like Uniswap pools are identified by tracking deployer contract events, while off-chain entities like Binance require behavioral inference. Raw on-chain data is considered commoditized, but the enrichment and labeling layer is not, and other analytics vendors regularly request access to Nansen's labels because they are difficult to recreate.

Nansen made a deliberate strategic decision to go vertical on traders and investors rather than remain a horizontal data company serving many use cases. The company observed users discovering signals on Nansen and then leaving the platform to execute trades elsewhere, which motivated building an integrated trading product. Nansen now describes itself as a full-stack on-chain agentic trading product. Web3 gaming and other non-trading use cases have not materialized as major segments despite earlier excitement.

The Nansen AI mobile app allows users to interact with an agent to explore on-chain data and execute trades within a single product. The agent accesses data in real time including transactions from seconds ago rather than relying on a historical training snapshot. It can answer questions about token buyers, smart money holdings, top holdings of specific funds like Defiance Capital, and Wintermute market-making activity. Nansen uses Privy for self-custodial embedded wallets and multiple DEX aggregators including Li-Fi, OKX DEX API, and Jupiter on Solana, and shows users in the UI which aggregator is fulfilling each transaction. The agent also integrates with X for social media feeds and is adding prediction market integrations. A vanilla LLM lacks native access to on-chain flows, cannot provide real-time information, cannot natively sign transactions, and lacks domain understanding of significance, and Nansen addresses these gaps through its proprietary data layer and agent tooling.

The AI trading agent can accept a user's strategy and back-test it against historical data such as the prior one or three months, which Nansen describes as a superpower for retail investors. Nansen Gym trains trading agents by replaying on-chain history as if it were happening in real time, with the environment sped up so agents do not wait the actual elapsed time. This kind of simulation capability has historically been available only to institutions like Renaissance Technologies and large hedge funds. Data harmonization across chains is a prerequisite because concepts like DEX trades are implemented differently on every blockchain and every DEX.

Nansen Smart Money 1.0 identifies addresses with the highest profit and loss over the past 30, 90, or 180 days. Smart Money 2.0 aims to predict which addresses are likely to make money in the future rather than looking backwards, with a target accuracy of 60 to 65 percent against a 50 percent random baseline and a goal of two to three times uplift in precision to identify addresses that will end in the top one percent of trading the following week. Smart Money 2.0 was described as weeks away from release. Nansen has decided not to build an AML or compliance product, describing it as a distraction from its mission of surfacing signal for crypto-native traders and investors.

Nansen reduced its subscription price from approximately 2,000 dollars per month to a flat 49 dollars per month, and made Solana data fully free including labels, trading, and SOL staking, subsidized through staking revenues and trading fees. The company chose broad access over institutional gate-keeping. The current mobile app requires users to manually approve each trade via biometric MFA, a design Nansen calls vibe trading. The company frames its longer-term trajectory using a trust ladder concept analogous to self-driving cars, where users gradually cede more control to agents. Nansen predicts that by 2028 or earlier the default way to invest will be through agents rather than picking individual tokens, and that all asset classes including commodities, real world assets, stocks, and fixed income will eventually come on-chain. Nansen's command line interface is open source, available via npm install nansen-cli, and can be given to an AI agent as a tool, with Nansen's stated goal being presence in every agentic trading stack including for hobbyists using open source tools.