Block Open-Sources Berd: Your New Desktop AI Agent Command Center

In a move that could significantly streamline how we interact with AI agents, Block, the tech giant behind Square and Cash App, has open-sourced Berd. This desktop application, originally developed internally, aims to provide a unified environment for employees to manage and utilize a variety of AI models, tools, and projects.

Berd isn't just another web interface; it's a locally installed graphical desktop application. This design choice is deliberate, allowing for deeper integration with local files, projects, and tools. Available under a permissive Apache 2.0 license on GitHub, Berd is accessible to everyone, allowing for free use, modification, and redistribution, even commercially. With builds for macOS, Windows, and Linux, Block is making a strong play for broad adoption.

What is Berd and Why Does it Matter?

Block positions Berd as a "daily AI work surface". Imagine a single hub where you can:

• Initiate chats with various AI agents.
• Attach files and folders for context.
• Select from different AI models and agents.
• Work within persistent projects that save your progress.
• Configure AI providers, skills, and extensions.
• Review past conversation history.
• Build automations to streamline repetitive tasks.

The core philosophy behind Berd is transparency and control. Unlike generic chatbot wrappers, Berd emphasizes making the operational state of your AI interactions visible. You'll always know which project, files, agent, model, and session state are actively influencing a conversation. Configuration, often buried in complex settings menus, is integrated directly into the workflow, ensuring that failures and states are plainly visible, not hidden behind a veneer of overly-friendly assistance.

Berd's genesis lies in a common problem within large organizations: fragmented AI tool experiences. Block employees were using powerful agents like Goose, Claude Code, and Codex, but managing them meant navigating multiple interfaces and configuration systems. Berd was the solution, offering "one consistent desktop application across models and harnesses." This unification is particularly valuable for enterprises looking to manage AI workflows efficiently and consistently.

Giving AI Agents Distinct Personalities

One of Berd's most innovative features is its approach to distinguishing between different AI agents. Instead of a sea of identical chat windows, Berd assigns recognizable visual identities to agents, using animated characters called "Gloopies." These aren't just for show; they represent the agent's role, skills, and tools, providing a visual shorthand for their capabilities. Personas like "Pushback" (for devil's advocate) or "Choosey" (for decision-making assistance) help users quickly identify and utilize the right agent for the task.

This visual differentiation is crucial as users increasingly juggle multiple specialized agents. It moves beyond mere aesthetics, making agents more identifiable and functional. Block emphasizes that these characters are designed to be "focused, capable, companionable", avoiding overly toy-like or distracting interfaces.

An Orchestration Layer, Not a New AI Model

Architecturally, Berd is not a new AI model or a runtime itself. It's built using web technologies (Tauri 2 and React 19) and acts as an orchestration layer. It communicates with agent harnesses like Block's own open-source Goose framework via the Agent Client Protocol (ACP). Goose, which runs as a "sidecar" process, handles the core AI loop, while Berd manages the user-facing aspects like projects, sessions, and configuration.

This separation is key. Berd doesn't lock you into a specific model provider. Instead, it allows users to configure and connect to the harnesses and providers they prefer. This model-agnostic approach, combined with Block's contribution of Goose to the Agentic AI Foundation, signals a commitment to open standards and interoperability in the AI space.

Key Differentiators and Enterprise Considerations

Berd stands out in a rapidly growing market of AI agent tools:

• Open Source and Free: Unlike many commercial offerings, Berd is free, with users only incurring costs for their chosen AI model providers.
• Multi-Harness Support: It works with various agent frameworks, not just one vendor's proprietary models.
• Local-First Data: Conversation history is stored locally, enhancing privacy and control.
• Visual Agent Identities: Distinctive avatars help users differentiate and manage multiple agents.
• Persistent Projects: Enables users to resume work without re-establishing context.
• Enterprise Customization: Offers "distribution seams" for organizations to integrate their own managed settings and tools.

For IT and security teams, the local-first data model is a significant advantage. Conversation history, credentials, and other sensitive data remain on the user's device by default. Telemetry is also opt-in and anonymized, excluding crucial information like prompts, files, and source code. While Berd offers customization for enterprise deployments, it's important to note that it's an endpoint application, not a fully centralized management plane.

The Future: From Single-User to Multiplayer

While Berd focuses on the individual desktop experience, Block's roadmap points towards integrating its capabilities into Buzz, their collaborative workspace. This integration aims to seamlessly transition from private, single-user agent work to collaborative, multiplayer workflows, all while maintaining context and continuity across devices, including mobile.

Berd's release is more than just a new tool; it's a statement about the future of AI interaction. By prioritizing an open, local-first, and user-controlled environment, Block is betting that the surrounding context and orchestration will become the stable foundation as AI models continue to evolve.