Turing TokenHub
Turing TokenHub is the model Gateway within Turing AgentCore. It unifies model vendors, protocols, and call governance behind Project, Environment, and API key boundaries.
TokenHub can also serve ordinary applications independently. When an agent uses it, TokenHub is the AgentCore model egress, not a parallel agent runtime.
What it covers
| Capability | Description | Docs |
|---|---|---|
| Model access | Access models from different vendors through one entry point | Model list |
| Protocol adaptation | Chat Completions, OpenAI Responses, and Anthropic Messages | Core APIs |
| Model capabilities | Embeddings, rerank, image, video, speech, and realtime APIs | Generation endpoints |
| Call governance | API keys, rate limits, timeouts, retries, fallbacks, and request tracing | Reliability |
| Token metering | Token counting, usage, and cost views | Usage and billing |
Relationship to AgentCore
| Scenario | TokenHub position |
|---|---|
| Ordinary application | Standalone model-access Gateway |
| Backend agent | Model-call egress for the agent loop |
| Managed Chat | Model-call layer used by the hosted conversation |
| Future Agent Runtime | Governed model egress inside Runtime |
Boundaries
- TokenHub does not retain durable agent state; that belongs to AgentCore LTM.
- TokenHub does not execute tools; tools run in the application, Tool Gateway, or Runtime.
- Protocol compatibility does not mean identical vendor parameters, context limits, or output semantics.
- Token metering supports platform usage and governance; it is not business cost accounting.
Current and future
Current docs describe existing model APIs, catalog, and governance. Future work may add clearer routing, policies, quotas, and cost controls while keeping the application boundary stable; these are not current commitments until their contracts are defined.