Getting started with MAGIC#
Introducing MAGIC (Multi-Agent Generative Investment Copilots): a dedicated team of AI-powered financial experts at your fingertips.
Designed to be as user-friendly and code-free as possible, MAGIC is a powerful AI tool that can answer any financial question you throw at it. It combines information from Internet research, validated datasets and SigTech’s knowledge base with strong analysis skills and a suite of Python tools to come up with a suitable solution to your query.
You can use MAGIC to write code, create data visualizations, and generate reports.
Find out how to get started with MAGIC and get familiar with its features.
Note: This is a beta version of the MAGIC platform, which is in the process of being tested before its official release. If you have any questions or feedback, please get in touch by emailing us at magic@sigtech.com.
Meet your agents#
MAGIC consists of multiple autonomous agents acting as AI experts. Each one plays a specialized role, carrying out specific tasks and working in collaboration to best answer your prompt. Here are some of the standard agents:
Planner (hidden agent): The project manager of the pack, it works behind the scenes to create a comprehensive plan that addresses your prompt and delegates tasks to different agents.
Internet Browser: Efficiently scours the web for information, analyzes the retrieved content, and identifies key points.
SigTech Quant: Performs financial quantitative analysis across asset classes by retrieving financial market data, creating strategies, and using trade simulation tools for backtesting—as well as generating financial charts and reports.
SigTech Copilot: Acts as your ‘no-code’ helper, retrieving documentation and Python code from SigTech’s knowledge base to help you handle data and build strategies within the SigTech platform.
Fed Analyst: Has access to the Federal Reserve FOMC documents.
As we continue to develop MAGIC, we also plan to create other specialist agents, as well as add to existing agents’ capabilities. Please note that access to particular agents can vary between users and organizations.
Agent settings#
You can view and configure the settings for each agent in the Agent settings section of the interface. Here, you can view agent information, as well as configure which agents to use in your current topic by toggling their ‘Active’ status.
You need at least one active agent in a topic to answer a prompt.
How to use MAGIC#
Your prompt is where the magic begins. Start interacting with MAGIC straightaway by typing a question, an instruction, or just a simple “Hi”. Example prompts include:
What are the most popular finance news stories today?
Show me Bitcoin returns since 2019
Can you chart US GDP YoY over the past 5 years?
I want to research the impact of the US presidential election on the returns of S&P. Where should I start?
Write some SigTech code for a basket strategy using UK government bonds
We have some example prompts in the platform to help you get started:
Continue the conversation#
The chat doesn’t end after the first message. If you want to keep the conversation going—for example, to learn more about something specific, switch up your question, or improve the answer given—you can follow up with another prompt.
Like a back-and-forth exchange in real life, MAGIC remembers the ongoing chat you’re having with it. You don’t have to repeat yourself or start from the beginning again, even if you’re returning to it days after. Pick up where you left off in the same chat window, and MAGIC will seamlessly refer back to any previous dialogue.
Each chat is stored as a Topic, which you can access in the Topics sidebar on the left-hand side.
@ an agent#
Want to draw on a particular agent’s expertise? No problem, you can ask a specific agent to delve further into your query. Just type ‘@’ in the prompt line to bring up a list of agents you can directly address. You can ‘@’ multiple agents in one go.
For example, if you know you want to create code to use in your SigTech research environment, you can direct your question to the Copilot agent, who specializes in generating code derived from SigTech’s framework. Or ask the Planner agent if you want to reword a response or ask about the background of a statement, as it oversees the entire process from research to results.
Note: You can only ‘@’ active agents. To activate an agent, toggle its ‘Active’ status in the Configuration menu.
Start a new chat#
Want to start afresh? Quickly launch a new chat by clicking the icon next to Topics at the top of the left sidebar. This wipes MAGIC’s slate clean. In your new conversation, it won’t have any recollection of previous chats, and it doesn’t hold on to past information.
Troubleshooting#
Like working with a junior quant researcher, MAGIC won’t always get things right on the first go. Luckily, it’s quick and easy to go back to the drawing board and ask your agents to redo or refine something if the initial response doesn’t match your expectations.
Try adjusting your prompt, such as rephrasing questions or changing approach, to get an answer that’s more in line with what you’re looking for. Read the section below for some suggestions on how to use MAGIC effectively.
If you’re still not satisfied with the answer you get, you can press the Thumbs down icon at the end of MAGIC’s response. This lets us know about the problem you’ve encountered, so we can continue to improve MAGIC.
If you’re experiencing lag or unresponsiveness, try clearing your site data (i.e. cookies and local storage) in your browser.
In some instances, MAGIC might try to render an image that does not display properly—this occurs when the image is not hosted on a SigTech domain. Please note this is by design, as it is a temporary safety feature until we have content scoring functionality in place.
Run into a problem or need more help with MAGIC? Get in touch at support@sigtech.com
Tips on using MAGIC#
The quality of MAGIC’s answers largely depends on how you phrase your questions. Here are a few tips to help you get the most out of MAGIC:
Be as straightforward and specific as possible: Try to avoid ambiguous or open-ended questions. The more precise and direct your instructions are, the better MAGIC’s answer will be.
Use conversational, natural language: Like a conversation with a human, it’s worth asking follow-up questions if you want to improve the answer or simply explore the topic further.
Provide necessary context and details: Share any relevant background information to help MAGIC generate a more useful response. Break down complicated questions into manageable chunks: For particularly long or complex prompts, it’s better to break it down into simpler, shorter parts to aid MAGIC’s understanding.
Start a new topic: Prior discussions within a chat can bias subsequent results. If you’re making a standalone or unrelated query, it’s best to start afresh with a new conversation.
Interact with specific agents: Make use of specific agents’ expertise by directly addressing them with ‘@’. For example, @Planner to clarify prior statements, or @ other agents to deal with specialist questions.
Ask for extra information: Get a better understanding of how MAGIC has reached an answer by asking it to provide evidence and references.
Features#
As well as the main chat window and the list of agents, the MAGIC interface has three other elements: Topics, Tasks, and Actions. Let’s get to know them.
Topics#
All your conversations with MAGIC are saved to the platform. Access your chat history via the Topics sidebar, where you can rename topics for ease of organization, delete chats, and revisit previous dialogues as far back as your first ever prompt. Topics are organized by date of creation.
Click on the three dots next to your chosen topic to edit the name or delete the chat from your history.
Note: Graphs and visualizations are archived after one week, so you may want to save them as images. If you revisit an old chat and the generated images have disappeared, you’ll need to run the prompt again.
Tasks#
You can view the tasks each agent undertakes as part of your query. Broken down step by step, it gives you a clear and complete overview of how MAGIC approaches your problem, from initially searching the Internet and knowledge bases to analyzing and summarizing its findings.
A loading symbol by the task means it is currently being undertaken, while a tick symbol indicates the task is done. Hover over each task to read it in full and see which agent worked on it.
Actions#
You can see the list of current and completed actions carried out by MAGIC in the lower right sidebar. Witness MAGIC in motion by seeing its current actions play out in real time. Once a current action is finished, it is moved to the completed section.
For a deeper dive into how MAGIC operates, you can expand each completed action to see the code powering it.
From ideas to implementation#
Ready to take MAGIC’s insights and results to the next step? Export your chat as a Jupyter Notebook (.ipynb) or JSON data, or share it with others via URL in read-only format. You can do this by clicking the icon in the top right-hand corner of the topic window. Otherwise, anything you’ve entered, as well as the resulting output, will remain only visible to you within the MAGIC platform.
All answers and related code (if applicable) can be copied as plain text using the copy and paste icon at the bottom of MAGIC’s output.
If you’re using the SigTech Copilot agent to write code, we recommend you to have the platform research environment open in another tab or window, so you can easily copy over and run any code snippets.