Who berd is for#
Developer teams running private AI agent workflows
Berd runs entirely on local infrastructure, making it suitable for teams that need to keep agent inputs and outputs off third-party servers. You configure your own model endpoints and run workflows without sending data through a managed cloud service.
Skip if:
If your team is comfortable routing agent work through a managed cloud platform and wants minimal infrastructure responsibility, a hosted service like CrewAI or Dify Cloud is a simpler starting point.
Enterprise teams building custom agent distributions
The repository's distribution seams allow organizations to overlay private agents, managed provider settings, optional companion tools, and signing infrastructure on top of the public berd build. Enterprise teams can maintain a private build without diverging from the upstream codebase.
Skip if:
If you need a managed enterprise tier with support contracts and SLA guarantees, berd's early stage and issues-only contribution model may not meet the requirement.
Developers testing or extending the Goose backend
Berd pins a specific Goose backend commit via a lockfile. Developers can point `GOOSE_BIN` to a local path to bypass the managed pinned checkout and test a custom or upstream Goose binary against the berd frontend.
Skip if:
If your interest is in the Goose backend itself rather than a desktop frontend for it, the upstream Goose project may be the more direct path.
The problem it solves#
Cloud-based AI agent platforms require your workflows to run on vendor infrastructure. Every agent call goes through a managed service you do not control, your model choices are constrained to what the platform supports, and costs accumulate per-seat or per-use at whatever rate the vendor sets. For teams with data locality requirements or strict governance rules, routing agent work through a commercial platform is often not an option.
The deeper issue is model lock-in. Most paid orchestration services are built around specific model providers, which means switching models often means switching platforms. Teams that need to run agents against private deployments, local models, or multiple providers simultaneously find that cloud-first tools do not fit the requirement.
How it solves it#
Model-agnostic runtime
Berd connects to the Goose backend via a local ACP WebSocket sidecar and works with any model endpoint you configure. You are not tied to a single provider, and switching models does not require switching platforms.
Enterprise distribution seams
Organizations can create enterprise distributions by supplying managed provider settings, private resources, and release infrastructure through built-in repository seams. Private material never appears in the public source tree, so the public and enterprise builds share the same codebase.
Bundled, self-contained desktop app
Tauri bundles the Goose backend as an external sidecar, producing a self-contained desktop application. The pinned Goose binary is tracked in `goose-backend.lock.json` so builds are reproducible and do not require a separate backend install.
Optional companion CLI seam
Berd includes a distribution seam for bundling an optional companion CLI as an app resource. The public build does not require a private CLI package; enterprise distributors can provide and package their own implementation while retaining the normal berd build and validation flow.
User-local experiment system
Berd provides a user-local experiment system for enabling unstable UI or workflow behavior. Experiments default on in dev builds and off in production builds, letting teams test new features without affecting stable users.
Strengths and trade-offs#
Strengths
- Apache-2.0 license, backed by BlockApache-2.0 is permissive: you can run it commercially, fork it, and distribute it without copyleft obligations. It is backed by Block (the company behind Square and Cash App), which signals sustained maintenance investment for a very new project.
- Desktop-native, no cloud account requiredBerd runs as a native desktop application via Tauri. There is no account to create and no data routed through a third-party cloud service. Agent workflows run locally, on your hardware, against model endpoints you control.
- Reproducible backend pinningThe Goose backend version is pinned via `goose-backend.lock.json`, so every team member builds against the same backend commit. This eliminates version drift between environments and makes the setup predictable across the team.
Trade-offs
- -No outside pull requests acceptedBerd's contribution model is issues-only: the team closes PRs from outside authorized collaborators automatically. If you need a feature or bug fix, you file an issue rather than submitting code. Fix timelines depend entirely on the core team at Block.
- -Build-from-source setup requiredThere is no pre-built binary download described in the README. Getting started requires running `just setup` and `just dev`, which installs pnpm dependencies, builds the vendored Goose SDK, and prepares the backend binary. This is a developer-oriented setup, not a one-click install.
- -Very new project with limited track recordThe repository was created in August 2026, with 223 stars and 21 open issues at listing time. There is not yet a documented production track record or community-sourced guidance for edge cases. Factor project maturity into your evaluation.
berd vs alternatives#
berd vs CrewAI
berd and CrewAI both target AI agent orchestration workflows. CrewAI is a commercial platform for defining and running multi-agent teams; berd is an Apache-2.0 desktop application that runs the agent runtime locally through the Goose backend.
| Feature | berd | CrewAI |
|---|---|---|
| License | Apache-2.0 | Commercial |
| Deployment | Self-hosted desktop app | Cloud managed |
| Model support | Any model endpoint | Platform-dependent |
| Data routing | Stays local | Through vendor servers |
berd is the stronger choice when data locality is a hard requirement or when your team needs to run agents against private or custom model endpoints. CrewAI is worth considering when you want a managed service with no infrastructure setup and its built-in agent coordination abstractions fit your workflow.
berd vs Dify Cloud
Dify Cloud is a managed cloud service for building AI workflows and chatbots with a visual interface. berd is a self-hosted desktop app that delegates to the local Goose backend, with a developer-oriented workflow and no GUI workflow builder.
| Feature | berd | Dify Cloud |
|---|---|---|
| License | Apache-2.0 | Commercial (cloud) |
| Deployment | Self-hosted desktop | Vendor-managed cloud |
| Interface | Desktop application | Web-based GUI |
| Target user | Developer teams | Developers and non-technical users |
berd is the right choice for teams that need agent workflows to run entirely on their own infrastructure. Dify Cloud is the better fit when your team prefers a managed service with a visual workflow builder and does not have data locality requirements.
Install and self-host#
Berd is built from source; clone the repository, then run setup to install dependencies and prepare the pinned Goose backend.
```bash
git clone https://github.com/block/berd
just setup
just dev
```What it's built on#
- Languages
- JavaScriptRustTypeScript
- Frameworks
- React
FAQ#
What AI models does berd support?
Berd is model-agnostic: it connects to the Goose backend via an ACP WebSocket sidecar and works with any model endpoint you configure. You are not limited to a specific provider's API.
Is berd free to use?
Yes. Berd is Apache-2.0 licensed, which means you can run it, fork it, and use it commercially without licensing fees. The self-hosted build requires no account or subscription.
How do I install berd?
Berd is built from source. Run just setup to install pnpm dependencies and prepare the pinned Goose backend, then just dev to start the development build. The README also documents just bundle for building a production desktop release.
Can I contribute code to berd?
Berd does not accept pull requests from outside authorized collaborators; outside PRs are closed automatically. The supported contribution path is opening a well-formed issue. The CONTRIBUTING.md file specifies what each issue type requires.
Who maintains berd?
Berd is built and maintained by a small team at Block, the company behind Square and Cash App. It is developed in the open and the repository is public, but the core team controls what changes merge.
Similar open-source tools#
rakazo
AI teammates you own: your keys, your model, your machine.
mex
Persistent memory and code graph for AI coding agents
hiring-agent
AI-powered resume evaluation and scoring tool
CocoIndex
Incremental data framework for AI agents.
Tolaria
Organize your notes with Markdown and Git integration
OpenCode
OpenCode is an open-source AI coding agent that assists developers in