08 AI Provider Expansion

Scope:

• LLM provider expansion
• Knowledge provider expansion
• Orchestration pipeline stability
• AI observability and security controls

A1 provider-matrix

• [x] Inventory current LLMProvider and KnowledgeProvider contract gaps
• [x] Define the multi-provider configuration matrix
• [x] Define the priority, fallback, and circuit breaking strategy matrix
• [x] Build a declarative model for provider capabilities

Acceptance:

• Adding a provider no longer requires runtime if/else stitching

A2 llm-provider-expansion

• [x] Add a second real LLMProvider adapter placeholder
• [x] Unify the streaming/non-streaming response contract
• [x] Unify the tool calling / function calling contract
• [x] Add provider-level timeout, retry, and error classification

Acceptance:

• The LLM provider no longer defaults to OpenAI as the only path

A3 knowledge-provider-expansion

• [x] Add a second real KnowledgeProvider adapter placeholder
• [x] Abstract index, delete, rebuild, and retrieval consistency semantics
• [x] Define tenant / knowledge base multi-tenancy mapping rules
• [x] Add provider switching regression tests

Acceptance:

• The WeKnora compatibility provider becomes a replaceable implementation, not the only one

A4 orchestration-hardening

• [x] Add a query orchestration scenario test matrix
• [x] Add a prompt builder contract test matrix
• [x] Add retrieval degradation path tests
• [x] Document AI fallback behavior

Acceptance:

• Orchestration layer behavior stays stable across provider switches

A5 ai-observability-and-policy

• [x] Add provider-dimension tracing/span labels
• [x] Add token / latency / error classification metrics
• [x] Reserve content safety and policy refusal hooks
• [x] Reserve prompt/version audit record interfaces

Acceptance:

• AI capabilities have the basic observability and policy control points needed before going live