Assistant: Refactor Model Providers

[samclark2015]

Addresses #10532.

Summary

This PR refactors the Assistant extension's model provider architecture to improve code organization, maintainability, and developer experience. The changes address multiple items from the cleanup checklist in #10532.

Major architectural changes:

• Separated provider classes into individual files: Broke apart the monolithic 1,749-line models.ts file into a clean directory structure under providers/
• Created robust base classes: Introduced ModelProvider abstract base class and VercelModelProvider for shared functionality across providers
• Improved error handling: New modelProviderErrors.ts with structured error classes for better debugging
• Centralized logging: Added modelProviderLogger.ts with consistent logging patterns across all providers
• Better type organization: Extracted shared types into modelProviderTypes.ts

Naming consistency improvements:

• Renamed getLanguageModels() → getModelProviders() for accurate terminology
• Renamed provider property → providerId for consistency across model providers
• Renamed name property → displayName to better reflect its purpose
• Updated all related types, tests, and comments

Provider organization:

providers/
├── base/
│   ├── modelProvider.ts              # Abstract base class
│   ├── vercelModelProvider.ts        # Vercel AI SDK base class
│   ├── modelProviderErrors.ts        # Error handling
│   ├── modelProviderLogger.ts        # Logging utilities
│   └── modelProviderTypes.ts         # Shared types
├── anthropic/                         # Anthropic providers
├── aws/                               # AWS Bedrock
├── azure/                             # Azure OpenAI
├── google/                            # Google AI & Vertex
├── mistral/                           # Mistral AI
├── ollama/                            # Ollama
├── openai/                            # OpenAI & compatible providers
├── openrouter/                        # OpenRouter
├── posit/                             # Posit Connect
├── snowflake/                         # Snowflake Cortex
├── test/                              # Test providers (echo, error)
└── index.ts                           # Central export hub

What's been accomplished from #10532:

• ✅ Separate provider classes in models.ts into separate files
• ✅ Use more accurate language (LanguageModel → ModelProvider)
• ✅ Make AILanguageModel abstract class more robust
  • ✅ Make providerName abstract (now providerId)
  • ✅ Reuse more code / reduce boilerplate
  • ✅ Support custom SDKs beyond Vercel AI SDK
• ✅ Consistent logging across all providers
• ✅ Add integration tests for each supported provider

Release Notes

New Features

• N/A

Bug Fixes

• N/A

QA Notes

@:assistant

This is a refactoring PR with no user-facing changes. All Assistant functionality should work exactly as before, just with better code organization.

Verification steps:

1. Launch Positron
2. Open the Assistant pane
3. Configure and test multiple model providers:
   • OpenAI: Standard OpenAI models
   • Anthropic: Claude models (both direct and via Vercel)
   • Azure: Azure OpenAI deployment
   • Ollama: Local models
   • Google: Gemini models
   • Snowflake: Cortex LLM
4. For each provider:
   • Verify models load correctly in the model selector
   • Start a conversation and verify responses work
   • Check that provider names display correctly in the UI
5. Switch between different models and providers
6. Verify error messages are clear if credentials are missing/invalid

Example test conversation:

# In console
import pandas as pd
df = pd.DataFrame({'a': [1, 2, 3], 'b': [4, 5, 6]})

Then ask the Assistant: "What is in the df variable?"

All provider configurations should continue to work without any behavioral changes, regressions, or errors.

[github-actions]

E2E Tests 🚀
This PR will run tests tagged with: @:critical @:assistant

^{readme  valid tags}

[Copilot]

Pull request overview

This PR refactors the model provider architecture in the Positron Assistant extension to improve code organization and maintainability. The changes move from a monolithic models.ts file to a structured provider-based architecture with better separation of concerns.

Key changes:

• Introduced an abstract ModelProvider base class that all providers extend
• Moved provider implementations to individual files under src/providers/
• Renamed classes from *LanguageModel to *ModelProvider for consistency
• Standardized provider lifecycle methods and error handling

Reviewed changes

Copilot reviewed 34 out of 34 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/tokens.ts	Updated imports and references to use new AnthropicModelProvider
src/test/snowflake.test.ts	Updated import path for Snowflake authentication utilities
src/test/openai.test.ts	Renamed OpenAILanguageModel to OpenAIModelProvider and updated imports
src/test/anthropic.test.ts	Renamed AnthropicLanguageModel to AnthropicModelProvider and added mock model info
src/pwb.ts	Changed autoconfigure result property from signedIn to configured
src/providers/test/errorProvider.ts	New test provider that always throws errors for testing error handling
src/providers/test/echoProvider.ts	New test provider that echoes back user input for testing
src/providers/snowflake/snowflakeProvider.ts	New Snowflake Cortex provider with credential refresh support
src/providers/snowflake/snowflakeAuth.ts	Updated import path for extension utilities
src/providers/posit/positProvider.ts	Moved and refactored Posit AI provider with OAuth support
src/providers/openrouter/openrouterProvider.ts	New OpenRouter provider implementation
src/providers/openai/openaiProvider.ts	Refactored OpenAI provider with comprehensive documentation
src/providers/openai/openaiCompatibleProvider.ts	New provider for OpenAI-compatible endpoints
src/providers/ollama/ollamaProvider.ts	New Ollama provider for local model execution
src/providers/mistral/mistralProvider.ts	New Mistral AI provider implementation
src/providers/index.ts	Central export file for all providers with factory functions
src/providers/google/vertexProvider.ts	New Google Vertex AI provider
src/providers/google/googleProvider.ts	New Google Gemini provider implementation
src/providers/base/vercelModelProvider.ts	Base class for Vercel AI SDK-based providers
src/providers/base/modelProviderTypes.ts	Type definitions for model provider interfaces
src/providers/base/modelProviderLogger.ts	Standardized logging utility for providers
src/providers/base/modelProviderErrors.ts	Error type hierarchy for provider operations
src/providers/base/modelProvider.ts	Abstract base class defining provider interface
src/providers/azure/azureProvider.ts	New Azure OpenAI Service provider
src/providers/aws/awsBedrockProvider.ts	AWS Bedrock provider with SSO support
src/providers/anthropic/anthropicVercelProvider.ts	Anthropic provider using Vercel AI SDK
src/providers/anthropic/anthropicProvider.ts	Anthropic provider using native SDK with detailed refactoring
src/participants.ts	Updated comment reference from EchoLanguageModel to EchoModelProvider
src/models.ts	Deleted monolithic file - functionality distributed to provider modules
src/extension.ts	Updated imports and references to use new provider structure
src/diagnostics.ts	Updated import for getLanguageModels function
src/copilot.ts	Changed autoconfigure property from signedIn to configured
src/constants.ts	Moved ModelCapabilities interface to modelProviderTypes.ts
src/config.ts	Updated imports and autoconfigure result handling

[melissa-barca]

I built the branch and did some light testing of different providers and everything is looking good. Let's get this merged! 💃