The biggest problem with powerful AI agents isn't their intelligence. It's the mess of tabs, apps, and lost context you need to manage them, a daily friction that drains productivity. Block, the company behind Square and Cash App, just open-sourced its internal fix: a free desktop application called Berd.
Available now on GitHub under an Apache 2.0 license, Berd is not another chatbot or model. It’s a “daily AI work surface” designed to orchestrate multiple AI agents, tools, and projects from a single, local-first environment according to VentureBeat. This signals a critical shift from chasing the 'best' model to building a usable, consistent work environment anyone can control.
How Berd Solves the Fragmented AI Workspace Problem
Inside Block, engineers hit a wall. They had capable agents like Goose, Claude Code, and Codex, but using them meant juggling different interfaces, configuration systems, and context management. “Working across them meant navigating different interfaces, configuration systems, and ways of managing context,” said Lucinda Bell from Block’s Technology Communications.
Berd consolidates that chaos. It’s a locally installed desktop app where users can start chats, attach files or folders, choose agents and models, and work inside persistent projects. The persistent project concept is key. Users can return to an established collection of files, instructions, and agent configurations instead of reconstructing context for every single task. This turns agentic work from a series of one-off prompts into a continuous, manageable workflow.
The design prioritizes accountability over convenience. According to its product spec, users should always know which project, files, agent, model, and provider are shaping a conversation. The interface exposes failures and loading states plainly. This is a deliberate rejection of the assistant-style friendliness that often obscures operational reality in other AI tools.
Meet the Gloopies: Why AI Agents Need Visual Identities
Berd tackles a subtle but critical UI problem: in a workspace with multiple specialized agents, how do you tell them apart before you even start a conversation? Its solution is a recognizable visual shorthand.
Instead of identical chat windows, Berd gives agents distinct animated avatars called “Gloopies.” Its public site showcases predefined personas like Berdy, Pushback, Choosey, and Copycat, each built for a different style of work. Pushback plays devil’s advocate on drafts. Choosey helps narrow decisions. Copycat learns to write in your style.
“The avatars make the agent recognizable. Its role, skills, and tools make it useful.”
This visual layer is functional. As workers juggle several agents, a distinct character serves as immediate identification for an otherwise invisible package of specific instructions, tools, and permissions. Block’s product specification guards against gimmickry, describing the intended personality as “focused, capable, companionable,” and explicitly rejecting toy-like presentation.
The Architecture: How Berd Acts as an Orchestrator, Not a Model
Architecturally, Berd is the conductor, not the musician. It’s built with Tauri 2 and React 19 as a lightweight desktop app. Under the hood, it communicates with Goose, Block’s open-source agent framework, through the Agent Client Protocol (ACP). Goose runs as a bundled “sidecar” process, handling the underlying AI agent loop. Berd handles everything on top: projects, sessions, context management, and the user interface.
This separation is core to its value. Agent harnesses like Goose, Claude Code, or Codex manage the loop between a model and its tools. Berd provides the consistent desktop environment around those harnesses. Users configure whichever providers and harnesses they want; it’s not locked to Block’s stack.
A key detail for enterprise evaluation: Berd doesn’t just pull the latest Goose build. Its build process pins a specific Goose backend version with a lockfile, verifies it, and packages it with the app. This creates a defined, stable boundary between the desktop application and its backend, preventing silent drift. For companies, predictability matters as much as features. This architectural rigor is a lesson in risk management that goes well beyond the typical security vulnerabilities seen in other AI platforms.
How Berd Stacks Up Against Codex, Claude Code, and Cursor
The market is flooded with AI coding tools, but Berd’s differentiation is stark. It’s not inventing the agent workspace concept but is making it open-source, multi-harness, and usable beyond pure software development.
Every major competitor is either tied to a single vendor’s models, priced as a subscription, or laser-focused on developers. Berd is none of those.
| Product | Maker | Pricing | Model Support | Distinguishing Features |
|---|---|---|---|---|
| Berd | Block | Free, open-source (Apache 2.0); users pay model costs | Any configured provider, across multiple harnesses | Desktop workspace over harnesses; local-first data; visual identities; designed for non-engineers |
| Codex app | OpenAI | Included with paid ChatGPT plans ($20-$200/month) | OpenAI models only | "Command center" for coding agents; parallel agent threads |
| Claude Code | Anthropic | Included with paid Claude plans ($20-$200/month) | Anthropic Claude models only | Deep autonomous coding; terminal, IDE, and desktop app |
| Cursor | Anysphere | Pro $20/month; Ultra $200/month | Multiple frontier models | AI-native editor; unified workspace with local/cloud handoffs |
Berd’s economic pitch is flexibility: “Some tasks need the big brain. Most don’t.” It lets users match the model to the job, potentially curbing the runaway costs associated with always using the most expensive frontier models. This is a pragmatic counterpoint to the growth-at-all-costs mindset driving models like those from the $40 billion revenue behemoth OpenAI.
What Enterprises Need to Know About Deploying and Governing Berd
For IT and security teams, Berd’s defaults are built for scrutiny. Conversation history stays on the user’s device in the local Goose session database. Credentials default to the OS keychain. Telemetry is disabled by default and, if opted into, excludes prompts, messages, files, source code, and credentials from its payload.
Local-first, however, does not mean fully offline. When you ask a configured model to work on context, that prompt and material are sent to the model’s provider. The data path and terms depend entirely on the provider you choose.
Block has created “distribution seams”: defined points where organizations can overlay managed provider settings, private agents, their own update channels, and signing infrastructure without touching the public source code. This allows for governed, custom enterprise distributions.
XOOMAR Analysis: There’s a major caveat. “Enterprise distribution” here means customized endpoint software, not a managed SaaS platform. The documentation does not describe central management features like single sign-on, admin policies, org-wide permissions, or audit logs. Companies can package and deploy Berd, but governing hundreds of installations after the fact is an unaddressed challenge. Furthermore, while the source is public, the repository does not accept outside pull requests. Block retains tight upstream control, directing external contributors to file bug reports. For any organization considering Berd as core infrastructure, this closed contribution model makes Block’s long-term maintenance commitment a critical question.
Is Berd Just a Stepping Stone to Block's Bigger Plan?
Berd is purpose-built for private, single-user work with privileged local access. But Block’s strategic endpoint is collaborative. That future is called Buzz, its open-source, Apache 2.0-licensed multiplayer workspace where humans and AI agents work side-by-side in shared channels.
Brad Axen framed the split as a question of trust boundaries: “A local agent may have access to your filesystem and credentials. An agent in a shared Buzz room has its own identity and access scoped through the relay.”
Buzz treats agents as members with their own cryptographic Nostr identities, not bots bolted onto a human chat. Every action is a signed event in a searchable audit log, a foundational approach for accountability as AI usage scales. The roadmap is clear: “Going forward, our focus is Buzz, and we plan to bring the best parts of Berd’s single-player experience into it.”
What to Watch: Berd’s release is Block planting a flag for open, user-controlled agent orchestration. The immediate value is a free, powerful tool to unify your AI workflow today. The larger signal is the direction: a future where work can begin privately on your desktop with Berd and become collaborative in Buzz without switching tools or rebuilding context. For now, Berd is a fully capable open-source offering. But its true legacy may be as the proving ground for the seamless solo-to-team agent experience Block is building next.
Why This Changes Everything
- It moves the AI focus from individual model capabilities to building usable, persistent workspaces for daily productivity.
- It offers a local-first, free solution that prioritizes user control, data privacy, and workflow continuity over cloud convenience.
- As an open-source tool from a major fintech company, it could set a new standard for how organizations and individuals orchestrate multiple AI agents.
Originally published on XOOMAR. For more news and analysis, visit XOOMAR.
Top comments (0)