Pattern Proposal: Model-Agnostic LLM Facade with Mock Fallback

Summary

Abstracts multi-vendor LLM invocation behind a unified interface with a deterministic mock fallback for offline and CI testing.

Candidate Pattern

• ID: model-agnostic-llm-facade-with-mock-fallback
• Classification: intelligence
• Maturity: established
• Type: implemented

Observed Signals

No specific signals listed.

Evidence

Supporting Evidence

• src/cas_runner/llm/litellm_client.py (LiteLLMClient) - Production implementation wrapping litellm.
• src/cas_runner/llm/mock_client.py (MockLLMClient) - Mock client enabling deterministic zero-cost unit and flow testing.

Scope Assessment

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

Rationale

• Intent: Allow agent flows to run seamlessly in local dev, offline CI, and cloud environments without vendor lock-in.
• Context: Underlies all LLM-powered components in cas_runner including AgenticMerger and synthesis steps.
• Force: Direct vendor SDK dependencies create vendor lock-in and test fragility
• Force: Automated CI testing cannot make costly or flaky external API calls

Traceability

Artifact source: proposed-patterns