Quick Facts
- Block released Berd on GitHub under an Apache 2.0 license with free builds for macOS, Windows and Linux.
- Berd stores conversation history locally on the user’s device and disables telemetry by default.
- Block does not accept external pull requests, limiting the open-source release to use and forking rather than community contribution.
Block, the fintech company behind Square and Cash App, has open-sourced Berd, a desktop application it built to give employees a single workspace for working with AI agents across different models and tools. The repository hit version 0.6.2 on Aug. 18 and lists 91 contributors across seven public releases.
The software grew from a real friction point inside Block. Engineers and non-technical staff alike had access to agents through tools like Goose, Claude Code and Codex, but each came with its own interface and configuration system. “We had capable agents through Goose, Claude Code and Codex, but working across them meant navigating different interfaces, configuration systems, and ways of managing context,” said Lucinda Bell, Block’s technology communications lead.
What Berd Does
Berd lets users start chats, attach files or folders, choose agents and models, work inside persistent projects and manage skills and extensions from one window. It connects to Block’s Goose agent framework through the Agent Client Protocol and ships with Goose as the default backend. Users can also configure Claude Code, Codex or local models through tools like Ollama.
The application is built with Tauri 2 and React 19. Tauri uses the operating system’s native web renderer rather than bundling a full browser, which produces smaller binaries and lower memory usage compared to Electron-based alternatives.
Block created animated agent personas called Gloopies, with predefined characters including Berdy, Pushback, Choosey, Copycat, Tinker and Wildcard. “The avatars make the agent recognizable. Its role, skills, and tools make it useful,” Block wrote in a company blog post.
Privacy and Data Model
Conversation history is stored in a local Goose session database on the user’s machine. Credentials default to the operating system keychain. Telemetry is off by default, and if a user opts in, Block collects only a random installation identifier and a limited set of usage events. Prompts, files, source code, credentials and IP addresses are excluded from the telemetry payload.
Brad Axen, Block’s head of AI capabilities, noted that local-first storage does not mean data never leaves the device. When a user sends a prompt to a configured model, that content goes to the model provider under whatever terms that provider requires. Organizations choosing a provider still own that contractual relationship.
Business Case and Limitations
Block is not positioning Berd as a commercial product today. Axen said the initial return on investment comes from Block’s own use of the software, which makes internal agent work more consistent and teaches the company how employees use agents in practice. “Supporting organizations that want to deploy agents at work is an area of interest for us,” Axen said, adding that Block will share more when there is more to share.
The Apache 2.0 license allows any company to use, modify and redistribute Berd commercially, including for internal deployment. However, Block does not accept external pull requests. Outside developers can fork the repository and modify it independently, but changes cannot be contributed back to the upstream codebase.
Jack Dorsey shared posts from Block employees about Berd after the release. Owen Jennings, a Block executive, described it as the multi-model and multi-harness agent builder he uses daily. The company’s $40 billion-plus market footprint gives the release immediate credibility among enterprise teams evaluating agent tooling.
This article was written by an AI agent. Spotted an error? Send a correction and we will fix it.