Turing AgentCore
Turing AgentCore is the platform layer for agent applications. It organizes how an agent runs, retains state, accesses tools, and is exposed to callers within shared Project and Environment boundaries.
Component relationship
Turing AgentCore
├── Agent API / Runtime (evolving)
├── Agent Gateway
│ ├── TokenHub Gateway: model access
│ └── Tool Gateway: tools and external services (evolving)
├── Agent State
│ └── LTM: durable cross-session state
└── Access surfaces
├── Managed Chat
├── CLI + Skills
└── Developer Platform
TokenHub is not a parallel agent platform. It is the model-call layer: it can serve ordinary applications independently and acts as the model egress for AgentCore agents.
Capability status
| Component | Current role | Status |
|---|---|---|
| TokenHub Gateway | Unified models, protocols, token usage, and call governance | Available today |
| LTM | Events, asynchronous derivation, Search, Context, and Space boundaries | Available today |
| Managed Chat | Managed memory orchestration and model calls in a hosted conversation | Available today |
| CLI + Skills | Local agents using Turing capabilities | Available today |
| Tool Gateway | A governed access target for tools and external services | In development |
| Agent API | Stable agent, session, and event invocation | In design |
| Runtime | Hosted execution, isolation, scaling, and runtime telemetry | Planned |
Two usage modes
TokenHub on its own
Ordinary applications, RAG services, or existing agent loops can call TokenHub protocol APIs without using AgentCore Runtime or LTM.
Composed AgentCore
An agent can call models through TokenHub, persist durable state through LTM, access external services through Tool Gateway, and eventually expose a unified invocation through Agent API or Runtime.
Related entry points
See Amazon Bedrock AgentCore overview. AWS similarly organizes Runtime, Memory, Gateway, Identity, and Observability as composable capabilities, while Turing resources and APIs evolve independently.