SwiftAgent

Overview

SwiftAgent is the smallest building block for building AI agents. Each agent is it's own autonomous unit, capable of accomplishing tasks by itself, but also capable of collaborating with other agents.

Actions

SwiftAgents are capable of having actions. These actions allow agents to interact with the external world. See the Actions section for more information about how to create actions and attach them to agents.

Actions

Memory

Memory is what gives agents their identity. Different kinds of memory allows for different agent functionalities, from semantic memory that allows agents to recall factual data, to episodic memory that allows agents to learn and evolve. Se tthe Memory section for more information about using Memory.

Memory

Persistence

Unlike traditional agent frameworks that rely on session-based memory storage, SwiftAgents introduce true end-to-end persistence—capturing the agent’s entire state in a single, restorable snapshot. This means every component, from its core configuration and behavioral instructions to its dynamic actions, self-descriptions, and accumulated memories, is preserved when saved. Where other systems might retain fragments (like short-term memory or transient session data), SwiftAgents enable seamless saving and loading of the agent’s holistic identity, including its "personality" (defined by instructions), operational logic, and lived experiences (memories). This allows agents to be paused, transferred, or redeployed across environments without losing context, ensuring continuity in long-term tasks, adaptive learning, or multi-step workflows. By persisting the full agent—not just its memory—SwiftAgents eliminate the need for manual state reconstruction, offering developers a self-contained, versionable entity that behaves consistently across interruptions, making it ideal for real-world applications where reliability and state integrity are critical.

Reasoning

Coming Soon!