Model Selection and Providers

Model Selection and Providers

CodingScaffold separates model recommendation from request routing.

Bootstrap Boundary

Model recommendation is available before any LLM is configured. CodingScaffold reads local project metadata, hardware facts, credential presence, and the prompt text. It then recommends a route. It does not proxy the prompt or call a provider.

Actual request routing happens later in the coding tool or an optional backend such as RouteLLM.

Recommendation

Use tools select-model when you want an explainable routing suggestion:

coding-scaffold tools select-model --target ~/dev/my-project \
  --prompt "Review this authentication refactor for security regressions."

The command does not call a model. It classifies the task and reports:

• prompt profile
• route: routine or heavy-lift
• provider
• model family
• model or deployment
• confidence
• reasons

Auto Mode

Use auto mode when a developer does not want to choose each time:

coding-scaffold tools select-model --target ~/dev/my-project --mode auto \
  --prompt "Fix this failing formatter test."

Auto mode still prints the decision so the user can challenge it.

Provider Detection

Provider detection checks:

• local runtimes such as Ollama, LM Studio, and llama-server
• local credential files in .coding-scaffold/.env.local
• project-local JSON credentials
• common cloud provider environment variables
• optional GitHub Copilot CLI status during explicit probe and doctor commands

Secrets are never printed.

Azure Guidance

For Azure OpenAI, use:

AZURE_OPENAI_API_KEY=
AZURE_OPENAI_ENDPOINT=
AZURE_OPENAI_DEPLOYMENT=

For Azure AI or Cognitive Services style endpoints, use:

AZURE_AI_API_KEY=
AZURE_AI_ENDPOINT=
AZURE_AI_MODEL=
AZURE_AI_MODEL_FAMILY=openai

If the endpoint serves Anthropic-family models, set:

AZURE_AI_MODEL_FAMILY=anthropic