- Load Module:
Import-Module ./PSAISuite.psd1 -Force - Run Benchmarks:
Import-Module ./PSAISuiteBenchmarks/PSAISuiteBenchmarks.psm1; Invoke-Benchmark -Models 'openai:gpt-4o','anthropic:claude-3-5-sonnet' - Validation:
Invoke-ScriptAnalyzer -Path ./
- Unified Interface: All new provider integrations MUST adhere to the standardized
-Messagesand-Modelparameter set. - Abstraction over Specificity: Avoid adding provider-specific parameters to
Invoke-ChatCompletion. Use the internal mapping logic to handle provider quirks. - Pipe-First Design: Ensure
Invoke-ChatCompletionconsistently accepts pipeline input for context. - Error Handling: Standardize error responses across providers so the calling agent/user receives consistent failure objects.
- Follow Verb-Noun naming conventions.
- Maintain the "Thin Wrapper" philosophy: minimize overhead between the user and the LLM endpoint.
- Ensure tab-completion for
-Modelis updated when new models are added toGet-ChatProviders.