The $10 trillion OpenBB Copilot validation

How BlackRock's Aladdin Copilot confirms OpenBB's vision for AI in finance

At the recent AI Engineering Summit, BlackRock unveiled its Aladdin Copilot, a platform remarkably similar to what we've built at OpenBB, but with a key difference.

While they've invested massive resources into building a closed system, we've created an open-source solution that achieves the same goals: multi-application support, seamless agent integration, explainable AI, and enterprise-grade security. Here's a deep dive into how the world's largest asset manager validated our approach to AI-powered financial workflows.

On Friday 21st, I attended the AI Engineering Summit from @swyx and Ben. This is arguably one of the best conferences available if you're working in the agentic space.

One presentation that particularly caught my attention was BlackRock's session about their Aladdin Copilot. While I can't share the presentation materials as they're conference-exclusive, what I saw left me incredibly bullish about OpenBB's direction.

Here's why: BlackRock's copilot is remarkably similar to what we've built at OpenBB - but with 1% of their resources and one major distinction.

Let's dive into the striking similarities first.

Not a single workflow, but multiple applications

BlackRock's Aladdin platform centers heavily on Portfolio - enabling users to handle portfolio construction, management, and monitoring.

In contrast, OpenBB's Workspace functions as an open playground. While it certainly handles portfolio management (as demonstrated in this example), it extends far beyond that. Our platform supports risk management, equity/crypto/macro research, ideation, ranking, client advisory, and even compliance workflows.

This versatility stems from our workspace architecture where users build on their own data. The flexible widget creation system can accommodate virtually any workflow - provided users have the necessary data.

Not a single chatting interface, the agent is on the side and is invoked when needed

This is a hill I'm willing to die on.

Analysts and PMs generally don't want a chat-only interface for their daily work. I wrote extensively about this eight months ago, and my conviction has only strengthened.

BlackRock appears to share this view. Their agentic copilot acts as a sidebar to the main interface, allowing users to query dashboard data and quickly validate information without disrupting their workflow.

It was genuinely shocking to see this in their demo—it looked remarkably similar to what we've had in OpenBB for over a year now. It's validating to see the world's largest asset manager (with 20,000 employees) arrive at the same conclusions we did.

Explainability

Both platforms prioritize data transparency. Every copilot response that references dashboard data highlights its source. This enables users to validate LLM outputs and trace information back to its origin, maintaining trust and accountability.

An example of how it can highlight a sentence or table level in an unstructured document:

An example of how it can highlight the widget origin that was used to answer to the prompt:

Secure environment first approach

This is non-negotiable in finance, where both data and prompts can be competitive advantages. Our president, Heidi Jonhson, recently detailed our approach to this in our on-prem announcement.

Architecture

While I can't share specifics from BlackRock's presentation, I can explain OpenBB's architecture, which appears to follow identical principles.

Our OpenBB Copilot acts as an orchestrator, gathering context from three main sources (in order of importance):

  1. In-context: Either attached files or explicitly referenced data widgets

  2. Dashboard: Data currently visible in the dashboard

  3. Product-wide: Connected to the workspace but not visible

This context exists in the form of widgets, and there can be thousands! In the case of BlackRock, they refer to this as Plugin Registry.

What is a widget?

In our system, a widget combines:

  • Data origin (API endpoint, static file, SQL query with DB connection, etc)

  • The parameters that can be modified to query a variation of the data

  • Metadata (title, description, category, sub-category, and source)

The metadata enables our copilot to identify and utilize appropriate widgets based on user prompts, by controlling the widget through its parameters.

You can think of these widgets effectively as tools that are rendered in our workspace. Therefore, our agent can call different widgets to retrieve the data it needs to reply more effectively to the user.

For a deeper dive into this architecture, check out my recent 10-minute presentation or the blog here.

Distinction

So what is the main distinction?

How Open we are.

We have an open-source data integration framework that enables any firm to bring any type of data into our product.

We have an open-source agentic framework that enables any firm to build its own agent (even one running locally).

And we intend to open source much more.

We believe in a future where each firm will build its tools on top of the most popular open source infra.

If you fall under that umbrella, reach out.