GOOSE PROJECTS (AI)
GOOSE DEVELOPMENT KIT (GDK)
Contrary to how it sounds, GDK is not a platform for developing geese, despite that being the only thing better than what it actually is.
Instead, GDK is an open-source SDK for building agentic applications that are model-agnostic, locally-runnable, and independent of any single provider. Built in Rust and powered by the Agent Communication Protocol (ACP), it provides the foundation for an open ecosystem of agents, clients, and tools. As part of the Linux Foundation's AAIF initiative, we're working with others in the industry to ensure the future of AI is built on open standards rather than closed platforms. Why? Because there's a war coming for who controls intelligence and unlike the wars for your attention, inbox, and social graph, this is one that we can win but only if we start now.
We built the Lightning Development Kit so that anyone could integrate bitcoin's Lightning network without asking permission. GDK is our effort to bring that same open-source ethos to AI, replete with important technical concepts like agent loops, tool calling, and context management that doesn't care if your model lives on a laptop, in the cloud, or in Narnia.
GOOSE REFERENCE CLIENT (GRC)
GRC is a first-party application that demonstrates best practices for using GDK by being a power tool for users. It doubles as a cheat code for those looking to build their own client and comes in two forms:
1) Goose Desktop is the familiar goose desktop application experience for power users who like to honk in windows instead of terminals.
2) Goose CLI lets users drive the agent directly from their terminal. This is for users who'd rather inject goose into a shell script, drop it into continuous integration, or wire it into whatever weird workflow they've built over the last decade. Goose CLI also uses the Rust API and demonstrates how to build on it. It's like a Zen garden for nerds and their agents.
MESH LLM
MESH LLM is a peer-to-peer inference network that allows individuals to contribute spare compute to run open-source AI models. By spreading inference across a network of participants, it lowers the hardware barrier to entry and makes open models accessible to people who don't have a mountain of GPUs sitting around.
AI PROTOCOL DEVELOPMENT
Agents are only as open as the protocols they run on. We work on two, or three if you count the one lurking and breathing heavily in our attic.
Model Context Protocol (MCP) connects agents to servers, giving them tools, data, and the ability to act. We were early to this, since goose was one of the first clients to support MCP-UI, which grew into MCP Apps. We also donated and maintain the Rust SDK for MCP.
Agent Client Protocol (ACP) connects clients to agents. One protocol, and any client (terminal, desktop, IDE) can drive any agent. We built a Streamable HTTP transport for ACP and co-lead the working group standardizing remote transports for the whole ecosystem.
