MAGIC Offering

MAGIC (standing for Multi-Agent Generative Investment Copilots) is our user-friendly, AI-driven tool, designed to expertly answer any financial question you throw at it.

It comprises a collaborative team of expert AI agents powered by large language models. Each agent is purpose-built to perform a particular task in financial research and analysis, meaning MAGIC can autonomously write code, use Python tools, call external APIs, or access public and private datasets on your behalf.

Best of all, you don’t need any coding knowledge to use MAGIC. All you need to do is enter a prompt–for example, “Give me the most popular finance news today”, “Show me Bitcoin returns since 2019”, or “Can you plot US GDP YoY over the past 5 years?”–and MAGIC will handle the rest.

1. First, we construct custom benchmarks to measure and optimize performance for example tasks in financial domains.
2. We engineer specialist knowledge and tools for the agent, enabling it to operate autonomously.
3. We augment the LLM with prompt engineering, adjusting its contextual information and fine-tuning behavior to enhance agent performance even further.
4. Finally, we adjust wider system processes within MAGIC to improve interactions with other agents.

MAGIC can be incredibly helpful in providing insights, ideas, and information. We ground the response with data provided external to the models to increase the reliability. However, as with any AI tool, you should take the results as guidance only and always exercise your independent judgment.If you want to see the code behind how MAGIC has reached a decision, you can view it in the Actions section of the interface.

Please refer to our terms and conditions for more information.

No, your data will not be shared with other users. Any data you’ve entered, as well as the resulting MAGIC output, will remain only visible to you within the MAGIC platform.

The results given by MAGIC are grounded by sourced data (i.e. from documents, computations, web pages, etc), which reduces the risk of hallucinations. 

The LLMs delegate any financial analyses to the SigTech Framework API, which removes hallucinations by executing code with SigTech market data.

The final process of collating the separate worker information also allows for reflection steps to remove inaccuracies.

Compared with other AI tools, MAGIC stands out with its specialist knowledge and data, high-quality quantitative insights, use of credible sources, and data-powered analysis. Each agent is specially built to deal with a particular aspect of financial research and analysis, making MAGIC an ideal tool for financial and economic professionals.

The MAGIC workflow is designed to enhance reliability, with tasks delegated to specialist agents to produce grounded, well-rounded results. While other LLMs are geared towards instant answers to quick queries, MAGIC is developed to facilitate complex problem-solving on a larger scale.

From receiving your prompt to coming up with a complex and detailed answer, MAGIC works in a matter of seconds. Your team of AI agents work collaboratively, carrying out their tasks to produce a solution to your query.

The planning and reviewing process is slightly slower than direct language model queries. However, this systematic process enhances the quality of MAGIC’s response.

Like working with a junior quant researcher, MAGIC won’t always get things right on the first go. But with MAGIC, it’s much faster to go back to your agents and ask them to redo or refine something.

MAGIC can be used for anything related to finance, such as data analysis, strategy development, real-time insights, and more.You can use MAGIC to ask for financial insights, track the historical performance of a single stock, give an update on the latest financial news, analyze current macroeconomic factors, or even help with using SigTech’s platform.Whether you’re a financial journalist, portfolio manager, quantitative analyst, independent investor, or simply someone who’s interested in finance, MAGIC can help get the answers you’re looking for.

If you have a particular question you’d like help with, get in touch with your technical account manager or contact support@sigtech.com. You can also refer to our Getting started guide for an overview of how to use the tool.

No, we do not use client prompts to train the AI models.