Pattern Proposal: Provider-Agnostic LLM Facade

Summary

Encapsulates LLM interactions behind a unified client interface to ensure provider portability and uniform call semantics across agent tasks.

Candidate Pattern

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

Observed Signals

No specific signals listed.

Evidence

Supporting Evidence

• src/cas_runner/llm/litellm_client.py (LiteLLMClient) - Provides consistent invocation wrapper around LiteLLM for agent steps.

Scope Assessment

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

Rationale

• Intent: Decouple agent business logic from specific model providers (OpenAI, Anthropic, Ollama, etc.) while standardizing call contracts.
• Context: Used across all runner steps and governance modules requiring LLM generation or structured output parsing.
• Force: Different model providers feature varying API schemas and error structures
• Force: Pipeline components need uniform invocation mechanics without vendor coupling

Traceability

Artifact source: proposed-patterns