AgentKitten: Swift package for provider-agnostic AI agents
FreeFreeFree tier
Inputs: textOutputs: text
About AgentKitten: Swift package for provider-agnostic AI agents
AgentKitten is a Swift package for building provider-agnostic AI agents on Apple platforms. It provides standard building blocks such as runtime tool permissions and hooks, context compaction, session KV store, validation loops, and auditable traces, allowing developers to create customizable agents without being tied to any specific LLM provider API. The package is designed to work with the Foundation Models framework on Apple platforms and aims to simplify agent development by offering reusable, modular components. Note that the project is evolving to complement Apple's official improvements to Foundation Models, with a focus on runtime tool permissions, approval gates, hooks, validation, and judge loops.
Key Features
Provider-agnostic abstraction over LLM APIs
Runtime tool permissions and approval gates
Context compaction for managing conversation history
Session KV store for persistent state
Validation and judge loops for agent behavior control
Support for hooks and middleware
Auditable traces for debugging and evaluation
Swift package designed for Apple platforms (iOS, macOS, etc.)
Pros & Cons
Pros
- Completely provider-agnostic – works with OpenAI, Anthropic, Google, etc.
- Provides essential agent infrastructure out of the box
- Designed for Apple ecosystem (Swift, Apple platforms)
- Open source and free to use
- Includes runtime security features like tool permissions and approval gates
- Offers auditable traces for transparency and debugging
Cons
- Limited to Apple platforms (iOS, macOS, etc.)
- Project is undergoing API churn as it aligns with Apple's Foundation Models framework
- Relatively new and may have a small community/ecosystem
- Documentation may be evolving and not yet comprehensive
Best For
Building customizable AI agents for Apple appsExperimenting with agent architectures and behaviorsPrototyping complex multi-tool agents with safety controlsEvaluating and debugging agent decision-making processesLeveraging Foundation Models framework with additional runtime controls
FAQ
What is AgentKitten?
AgentKitten is an open-source Swift package that provides standard building blocks for creating provider-agnostic AI agents on Apple platforms, abstracting away concrete LLM provider APIs.
Is AgentKitten free to use?
Yes, AgentKitten is open source and free to use. The source code is available on GitHub under a license (see LICENSE file).
What platforms does AgentKitten support?
AgentKitten is designed for Apple platforms, including iOS, macOS, watchOS, and tvOS, as it is built with Swift and integrates with the Foundation Models framework.
How does AgentKitten relate to Apple's Foundation Models framework?
AgentKitten originally provided an abstraction layer over different LLM providers. With Apple's introduction of the Foundation Models framework at WWDC 2026, AgentKitten is evolving to complement and extend that framework, focusing on features like runtime tool permissions, approval gates, validation loops, and judge loops.