Pattern Proposal: Dual-Mode LLM Provider Abstraction

Summary

Isolates model invocation behind a provider client interface pairing a multi-model gateway (LiteLLM) with a fully deterministic mock client.

Candidate Pattern

• ID: dual-mode-llm-provider-abstraction
• Classification: intelligence
• Maturity: established
• Type: implemented

Observed Signals

No specific signals listed.

Evidence

Supporting Evidence

• src/cas_runner/llm/litellm_client.py (LiteLLMClient) - Production LLM wrapper handling API calls via LiteLLM.
• src/cas_runner/llm/mock_client.py (MockLLMClient) - Mock client delivering canned deterministic responses for test runs.

Scope Assessment

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

Rationale

• Intent: Decouple agent business logic from specific cloud LLM APIs and allow zero-cost deterministic offline testing.
• Context: Instantiated by runners, analyzers, and mergers to execute LLM inferences.
• Force: Directly binding agents to a specific LLM SDK makes multi-cloud deployment or model migration arduous.
• Force: CI pipelines fail or accrue massive costs when executing integration tests against live LLMs.

Traceability

Artifact source: proposed-patterns