Pattern Proposal: Provider-Agnostic LLM Gateway with Dual Client Contract

Summary

Abstracts language model calls behind a standardized client interface with production and mock implementations.

Candidate Pattern

• ID: provider-agnostic-llm-client
• Classification: intelligence
• Maturity: established
• Type: implemented

Observed Signals

No specific signals listed.

Evidence

Supporting Evidence

• src/cas_runner/llm/litellm_client.py (LiteLLMClient) - Standardized runtime provider client.
• src/cas_runner/llm/mock_client.py (MockLLMClient) - Deterministic test double for model responses.

Scope Assessment

• Evidence Strength: high
• Confidence: high
• Applies to: ideas-to-life-cas

Rationale

• Intent: Isolate agent orchestration logic from model provider SDK churn and support deterministic offline testing.
• Context: Agent components like AgenticMerger require model completions, but testing requires offline predictability.
• Force: Model vendor SDK APIs change frequently
• Force: Unit and CI test suites cannot rely on live LLM API keys or incur token costs

Traceability

Artifact source: proposed-patterns