For Japanese version, see IModelProvider-jp.md.

IModelProvider (Interface Specification)

1. Responsibility Boundary

IModelProvider is the intelligence execution-engine boundary that consumes normalized messages and performs concrete model inference.

• Role:

Provide full generation, streaming generation, and direct Q&A while abstracting provider-specific transport differences.

• Non-role:

Model selection belongs to IModelVectorRouter; provider lifecycle ownership remains in IProvider.

2. Contract Signature

namespace AIKernel.Abstractions.Providers;\n\npublic interface IModelProvider : IProvider\n{\n    Task<string> GenerateAsync(IReadOnlyList<IModelMessage> messages, CancellationToken cancellationToken = default);\n    Task StreamGenerateAsync(IReadOnlyList<IModelMessage> messages, Func<string, Task> onChunk, CancellationToken cancellationToken = default);\n    Task<string> AnswerAsync(string question, string? context = null, CancellationToken cancellationToken = default);\n}

3. Related Use Cases

• UC-02 Structure phase execution:

Serves as the execution body beneath reasoning construction.

• UC-04 Generation and output polishing:

Produces source outputs for downstream polishing/rendering.

4. Governance & Determinism

• Determinism contribution:

Implementations should maximize replay stability through fixed inference settings where possible.

• Fail-Closed:

Safety violations, token-limit breaches, or execution faults should fail explicitly instead of returning partial answers.

• Isolation compliance:

Do not inject implicit context beyond supplied messages.

5. Implementation Notes

• Transport abstraction:

Hide REST/gRPC/local runtime differences behind this contract.

• Extensibility:

Evolve multimodal support in sync with IModelMessage contract extensions.

Changelog

• v0.0.0 / v0.0.0.0: Initial draft
• v0.0.1 (2026-05-06): Version upgrade aligned with documentation guidelines