Two MIT students want to put an AI research agent inside your brokerage account. The pitch is straightforward enough: connect your Robinhood or Fidelity portfolio, and algorithms will scan earnings reports, SEC filings, and price movements to flag trades you might want to make.
Ekpa, backed by Y Combinator in the Summer 2026 batch, has built what founders Anant Asthana and Yuga Patel describe as an "AI-native investment platform." The two-person Boston operation connects to 32 brokerages spanning North America, Europe, India, and Australia through read-only integrations, meaning the software can see what you own but cannot execute trades on your behalf.
The service monitors portfolios for what the company calls "concrete moves." An example on Ekpa's homepage suggests buying AMD based on momentum indicators. Another recommends trimming a specific number of Nvidia shares, with snippets from Reuters, Bloomberg, and CNBC attached to tickers users already hold. A chat interface fields investment questions using the portfolio data, though conversations are stored locally in browsers rather than on company servers.
Whether this constitutes genuinely useful investment advice or simply repackages widely available information remains an open question. The founders claim their AI trading systems have beaten the S&P 500 "over short time periods," citing returns exceeding 1.5% in a single week. Such short-term performance figures are difficult to evaluate without additional context about risk, position sizing, or the number of attempts required to generate such results. One-week returns in financial markets can reflect luck as much as skill.
Ekpa added a simulation feature called Portfolio Lab in August 2026, according to a company LinkedIn post. The tool runs Monte Carlo analyses using resampled market data since 2000, producing 80% outcome bands for hypothetical scenarios. Users can model expenses and portfolio adjustments without moving actual money. As of mid-September, counters displayed on the homepage indicated the platform had processed more than 6,500 analyses and connected over $7 million in assets, though these figures are self-reported and unverified.
The startup positions itself carefully in regulatory terms. Its terms of use, effective late July, specify that Ekpa operates as a research and education product, not a broker-dealer or investment adviser. SnapTrade handles brokerage connections by routing users to each institution's native login page, so credentials never pass through Ekpa's systems. Passwords are stored as salted scrypt hashes. Disconnecting a brokerage immediately deletes imported holdings, the company says. Stripe processes payments once the current free early-access period ends, though the company has not announced pricing or timing for that transition.
The space Ekpa has entered is increasingly crowded. Composer, now owned by SoFi after an acquisition finalized in June, offers AI-powered strategy creation with direct execution capabilities. Reflexivity, previously known as Toggle AI, introduced a Microsoft 365 Copilot plugin around the same time. PortfolioPilot, run by SEC-registered adviser Global Predictions, reported 40,000 users and $30 billion on its platform as of November 2025. Institutional players are also experimenting: Lemma Trading markets research agents designed to generate trade ideas and then stress-test them for trading desks.
A report from the Cambridge Centre for Alternative Finance published in April noted that roughly one-fifth of surveyed financial-services firms had deployed some form of agentic AI, citing data from NVIDIA. The technology's appeal is obvious. Markets generate torrents of data, and parsing it manually grows harder as retail investors accumulate positions across multiple accounts and asset classes.
Whether algorithms can reliably synthesize that information into profitable actions is less clear. Markets tend to punish patterns once they become widely known, and automated investment tools have historically struggled to consistently generate outperformance after accounting for fees and market conditions. Robo-advisers enjoyed a surge of enthusiasm a decade ago, then settled into a narrower role as low-cost portfolio rebalancing tools rather than alpha generators.
Ekpa's founders are betting that newer AI models can do better. Asthana has STEM competition awards on his resume; Patel authored the company's Y Combinator launch post in July. Y Combinator typically invests $500,000 in each accepted startup under its standard deal. Ankit Gupta is listed as the primary partner working with the company, according to the accelerator's directory.
The platform supports brokerages including Schwab, Vanguard, Trading 212, DEGIRO, Questrade, Wealthsimple, Zerodha, and Stake. Email alerts notify users of significant moves in stocks they own or broader market shifts, with an optional morning digest. The company promises users will receive advance notice before pricing kicks in, though no timeline has been specified.
For now, the product remains free and the team remains small. Two people, a handful of months since launch, and a growing list of integrations. It is the sort of setup that defines early-stage software startups: ambitious technical scope, uncertain business model, and a market that may or may not exist at scale. The next year will clarify whether investors actually want an AI agent combing through their portfolios, or whether the appeal fades once the novelty wears off.
