Scalar Field: The agentic trading desk

Hi all,

We’re Aman and Ramakant, founders of Scalar Field.

I previously worked as a trader at Tower Research and Goldman Sachs. Ramakant led engineering teams at Microsoft. We’ve both spent our careers at the intersection of finance and technology, and we’re now building the infrastructure financial agents need to operate in markets.

TL;DR: Scalar Field lets you turn an investment idea into research, a backtest, and a live trading agent in one place. Agents can use LLMs for research and reasoning, then run strategies through an event-driven execution layer with approximately 300 ms event-to-trade latency.

https://youtu.be/Uhd-RtQ1XjM

The problem

Over the next few years, agents will not just help people research markets. They will actively participate in them.

They will monitor news, parse filings, analyze data, test hypotheses, manage portfolios, and execute trades.

But today’s financial infrastructure was not built for this.

Market data is fragmented across providers. Research and backtesting happen in separate systems. Execution requires another set of integrations. Most financial platforms are dashboards designed for humans, not infrastructure designed for agents.

There is also a more fundamental problem: most AI-agent frameworks assume every action should begin with an LLM call.

An event occurs. The agent sends context to a model. The model reasons about it, chooses a tool, and eventually places a trade.

That is often the wrong modality for trading agents.

LLM calls are costly and introduce significant latency. A trading agent cannot call a language model every time a price changes, a signal fires, or a portfolio needs to be rebalanced.

LLMs are useful for interpreting unstructured information, researching ideas, and developing strategies. They should not have to sit in the critical path of every trade.

What Scalar Field does

Scalar Field separates the agent’s reasoning layer from its execution layer.

Agents can use LLMs where intelligence and interpretation are valuable. They can then convert those decisions into deterministic, event-driven strategies that continuously monitor markets and execute without another LLM call at every step.

Scalar Field handles:

• Financial data and research compute
• Python-based strategy development and backtesting
• Event and schedule-based strategy triggers
• Persistent agent state
• Portfolio allocation and position tracking
• Risk and execution checks
• Order routing and reconciliation
• Real-time performance and NAV tracking

This lets users go from thesis to backtest to live portfolio without stitching together data providers, compute infrastructure, agent frameworks, and brokerage APIs.

Where agents can trade

Scalar Field currently supports:

• Robinhood — US equities and ETFs
• Public.com — US equities and options
• Polymarket — prediction markets
• Jupiter DEX — Solana tokens and tokenized assets
• Alpaca — US equities and options

We also support paper trading through Alpaca and Polymarket, so users can test agents before allocating real capital.

Hyperliquid support for crypto perpetual futures and spot is coming soon.

What you can build

For example, you could build an agent that:

• Tracks which public companies are increasing AI capital expenditure, then continuously rebalances into the companies investing most aggressively in data centres, chips, and AI infrastructure.
• Monitors insider purchases across thousands of companies, tests whether purchases by CEOs or directors predict future returns, and automatically trades only when the historical signal is strong enough.
• Reacts to market-moving events such as earnings surprises, inflation data, interest-rate decisions, breaking news, or sudden changes in prediction-market probabilities.
• Watches every live sports game and trades when the market overreacts, for example, when the team that started as the favourite falls behind early but its estimated win probability begins to recover.
• Runs a simple deterministic rule, such as buying a stock when a predefined signal crosses a threshold, without making an LLM call at all.
• Or goes much deeper: trains a transformer on historical market data, uses the model to generate signals in real time, and automatically executes trades when those signals meet the strategy’s risk and confidence requirements.

Our long-term vision is agentic portfolios: portfolios created, managed, and continuously rebalanced by agents around any thesis, narrative, event, or market signal.

We believe agents will become some of the largest participants in financial markets. But that will not happen by simply connecting a generic LLM-agent framework to a brokerage API.

Trading agents need infrastructure that combines intelligence with financial data, event-driven compute, persistent state, portfolio management, risk controls, and low-latency execution.

That is what we’re building with Scalar Field.

Asks

1. Try Scalar Field and tell us what breaks, what is missing, and what trading agent you would build first.
2. Introduce us to quantitative traders, developers, and investment teams experimenting with AI-driven research, portfolio management, or automated execution.

You can try it at scalarfield.io.