goose OSS Roadmap (Feb–Apr 2026)

[DOsinga]

🪶 goose OSS Roadmap (Feb–Apr 2026)

Welcome to the goose OSS roadmap for February → April 2026. This document captures the major themes we’re focused on and the bets we’re making to move goose from a great AI assistant to a local-first, extensible agent platform with a strong out-of-the-box experience.

⸻

🌍 Open Models (Local-first, Sovereign by Default)

goose becomes the premier agent for open source AI, prioritizing local inference and private, sovereign workflows. We'll build inference and model download directly in the app, sidestepping the need for external runtimes. We'll optimize the agent loop, tool-calling and prompts to work well with smaller open models (local where possible, hosted otherwise). We'll also explore a peer-to-peer compute ecosystem so users can share or tap into idle capacity.

We plan to:

• 📦 Ship built-in inference and open model downloads directly in goose (no external runtimes required)
• 🧠 Tune prompts and tool-calling for small and open models (tighter schemas, fewer turns, predictable behavior), measured with an open source model “gym” to help optimize
• 🔁 Optimize the agent loop for local constraints (latency-aware planning, bounded retries, graceful degradation)
• 🌐 Explore optional peer-to-peer compute sharing with explicit opt-in, trust, and resource controls, explore payment for offering spare compute to pool with a application wallet

⸻

🧩 Apps (Vibe-Coded, Composable, and Deeply Integrated, real world commerce)

goose follows the MCP standard and implements MCP Apps as the foundation for interactive extensions. Separately, goose has pioneered support for vibe-coded apps, which are not part of the MCP standard (yet) but build on the same underlying ideas. At this point, goose is running ahead of the standard: the goal is no longer to prove that apps work, but to make them more powerful, composable, and deeply integrated with goose itself.

Vibe-coded apps should be able to connect to multiple MCP servers, define their own tools, and participate directly in goose workflows. If you build a todo app, goose should be able to list items, add new ones, and immediately reflect those changes in the app UI. Over time, apps should feel like native goose capabilities that collaborate with each other rather than sitting alongside the core.

We plan to:

• 🧩 Let apps connect to multiple MCP servers and compose their capabilities into a single app
• 🛠️ Allow apps to define their own tools that goose and other agents can call directly
• 🔄 Enable tight, live integration between apps and agents, keeping app state and agent actions in sync
• 🪶 Move vibe-coded apps closer to core goose, enabling apps to collaborate and extend each other
• Agentic commerce goes hand in hand with MCP apps. Make goose a great reference platform for commerce apps and sharing, including specs like UCP, wallets and payment

⸻

⚡ Out-of-the-Box Experience (Batteries Included)

goose should work exceptionally well out of the box. AI is no longer just for programmers, and a new user should be able to install goose on a supported OS and immediately get useful work done — without understanding MCP servers, providers, models, or configuration files.

The out-of-the-box experience should prioritize clarity, stability, and trust. Customization should remain possible, but never required to be productive.

We plan to:

• 🧰 Minimize setup and configuration so common workflows work immediately
• 🧠 Ship strong defaults for models and providers that behave predictably
• 🚨 Surface failures clearly with actionable guidance when something is missing or broken
• 🧭 Ensure consistent behavior across macOS, Windows, and Linux

⸻

🧠 Meta-Agent Orchestration (Many Agents, One Workflow)

goose should evolve beyond a single chat agent into a meta-agent orchestrator that manages sessions, workflows, recipes, and multiple subagents running in parallel. The UI is already moving in this direction with initial support for multiple sessions — and the broader ecosystem is converging on “multi-agent done sensibly.”

We plan to:

• 🧵 Make recipes first-class and composable, supporting subrecipes and parallel subagents
• 🗂️ Add clear task and progress tracking so multi-agent work stays understandable and debuggable
• 🧰 Define a concrete execution model for agents (including isolation and optional local containers)
• 🧭 Introduce explicit project types and workflows to make orchestration usable beyond power users

⸻

🪟 UI & Interaction (Composable, Not Prescriptive)

As goose expands beyond a single chat workflow, the UI needs to adapt to different ways of working. Rather than a fixed layout, goose should provide composable UI primitives that support richer interactions without sacrificing clarity.

We plan to:

• 🧩 Build UI primitives for apps and sidecar panels that can mount alongside conversations and share context
• ✍️ Improve the chat input to handle structured text, code, files, and commands naturally
• 🧭 Keep the default experience simple, while allowing layout and navigation to evolve as workflows get more complex

⸻

🔌 Protocol (Clear Boundaries Between Client, Server, Extensions)

To support goose as a platform rather than a single app, we need a stable protocol boundary between clients, servers, and extensions.

• MCP is the protocol for interacting with extensions.
• ACP is a strong contender for agent-to-client communication.
• A client–server architecture (even locally) gives us portability and experimentation without expanding the supported surface area endlessly.

The current goosed protocol is ad hoc and too broad to be fully captured by pure ACP, so we’ll likely need targeted extensions or layering.

We plan to:

• 🧱 Define a small, stable protocol surface with a clear line between supported and experimental behavior
• 🔌 Standardize a client/server architecture (even locally) and define how goose extensions and providers communicate over ACP
• 📱 Prove the protocol by building additional clients (e.g. TUI, mobile) and provider integrations on top of it

⸻

[abh3po]

Love the local-first approach have been waiting for it for quite some time, will check it out soon!

[JERRYFROMKENYA]

This is brilliant.

[Benoss]

Excited with the UI interactions, being able to ask structured questions/answers to the user instead of just prompt is I think going to change the way we see charbots

[bioinfornatics]

Did you plan to change sacp to agent-context-protocol, to allow that 1 agent own multiple mode?

[debs-obrien]

Can we also work on making it easier to update goose. unzipping and dragging and dropping into my applications folder makes me not want to update. its too many steps. So I don't. Then I end up with bugs that are already fixed and then I report them and waste peoples time asking for help when the answer was to just update but the process of updating is just too tedious. i just want to restart to update like VS Code does for example.

[u35tpus]

Please review the goose bug fix pull request #6885, this is serious bug for bedrock users

[mottobiz]

I'm originally a Windsurf user but have become a big fan of Goose already. I think"Meta-Agent Orchestration" is something I look forward to... I love the energy level of Goose users and the support team here 💪. Honestly I am still trying to learn how to make the best of Goose (Recipes especially 🥸). Would love to ship my first project entirely with Goose.

Grateful to have come across Goose and the team who are genuinely building everything with a user first interface. Thanks everyone. Would appreciate any tips to get me a head start ✨