Foundry

Foundry: The Training and Evaluation Infrastructure for Web-Native Agents

What is Foundry?

Foundry is a platform designed to facilitate the development and evaluation of AI agents operating within web environments. It addresses the challenges of building reliable AI agents that can navigate and interact with real websites effectively.

How does Foundry work?

Foundry provides deterministic environments, state-based evaluation, and informed data collection to create a robust feedback loop for agent improvement.

• Deterministic Environments: Guarantees consistent website versioning for reliable testing.
• State-Based Evaluation: Allows defining custom evaluation criteria and reward functions using structured state JSON.
• Informed Data Collection: Captures demonstration data for behavioral cloning and on-policy reinforcement learning when failures occur.

Why is Foundry important?

With over 60% of global knowledge work mediated through the browser, agents mastering this environment can unlock substantial economic opportunities across support, sales, and internal operations. Foundry offers a purpose-built infrastructure to enable the creation of these agents.

Where can I use Foundry?

Foundry can be used by researchers and engineers to build and test AI agents for web automation, ensuring they perform reliably under realistic conditions. It's particularly useful for:

• Automating workflows.
• Navigating SaaS tools.
• Searching databases.

Best way to get started with Foundry?

To get started with Foundry, reach out to the team for more information and explore opportunities to collaborate or join their team.