Skip to content

Alpha Berkeley Framework v0.7.6 - Provider Registry System

Choose a tag to compare

View all tags
@thellert thellert released this 30 Oct 19:05
· 1606 commits to main since this release
v0.7.6
5898b63

Alpha Berkeley Framework - Latest Release (v0.7.6)

🚀 Architecture Enhancement Release - Provider Registry System with extensible AI provider management.

What's New in v0.7.6

🏗️ Provider Registry System

Centralized Provider Management

  • Registry Integration: AI providers now managed through the same registry system as capabilities, services, and other components
  • Single Source of Truth: All provider logic centralized in provider adapter classes
  • Lazy Loading: Providers loaded on-demand with proper dependency management
  • Extensibility: Applications can register custom providers (Azure OpenAI, institutional AI services, etc.)

Provider Adapter Architecture

  • New BaseProvider Abstract Class: Defines consistent interface for all AI providers
  • Five Framework Providers: Anthropic, OpenAI, Google, Ollama, CBORG adapters implemented
  • Provider Metadata: Simple class attributes define requirements (API key, base URL, proxy support)
  • Health Check System: Each provider implements connectivity testing with charge-aware logic

Simplified Registration

  • Minimal Registration: Just module_path and class_name required
  • Metadata Introspection: Registry reads provider metadata from class attributes
  • 60% Less Code: Provider registration reduced from 8-9 lines to 2 lines per provider
  • No Duplication: Single source of truth eliminates metadata duplication

Code Reduction & Maintainability

  • ~860 Lines Removed: Eliminated provider-specific logic from factory.py, completion.py, health_cmd.py
  • ~290 Lines Saved: Factory module reduced from 472 to 206 lines (-56%)
  • ~298 Lines Saved: Completion module streamlined significantly
  • ~290 Lines Saved: Health check module simplified
  • Single Implementation: Both PydanticAI models and direct completions use same adapter code

🔧 Critical Features Preserved

All existing functionality maintained:

  • ✅ HTTP Proxy Support: HTTP_PROXY environment variable for enterprise deployments
  • ✅ Ollama Fallback Logic: Automatic localhost ↔ host.containers.internal switching
  • ✅ Timeout Configuration: Per-provider and per-request timeouts
  • ✅ Environment Variables: ${VAR_NAME} resolution in config files
  • ✅ Extended Thinking: Anthropic and Google thinking modes fully supported
  • ✅ TypedDict Support: Automatic conversion to Pydantic models
  • ✅ Charge-Aware Health Checks: Format validation for paid APIs (Anthropic, Google)

🎯 Developer Experience Improvements

Custom Provider Registration

Applications can now register custom AI providers:

from framework.registry import ProviderRegistration
from framework.models.providers.base import BaseProvider

# Define provider adapter
class AzureOpenAIProviderAdapter(BaseProvider):
    name = "azure_openai"
    requires_api_key = True
    requires_base_url = True
    # ... implementation ...

# Register in application registry
return extend_framework_registry(
    providers=[
        ProviderRegistration(
            module_path="my_app.providers.azure",
            class_name="AzureOpenAIProviderAdapter"
        )
    ]
)

Common use cases:

  • Azure OpenAI enterprise deployments
  • Institutional AI services (Stanford AI Playground, national lab endpoints)
  • Commercial providers (Cohere, Mistral AI, Together AI)
  • Self-hosted models with OpenAI-compatible APIs

Provider Discovery

from framework.registry import get_registry

registry = get_registry()
providers = registry.list_providers()  # ['anthropic', 'openai', 'google', 'ollama', 'cborg']
provider = registry.get_provider('anthropic')  # Get provider class

🧪 Testing Infrastructure

pytest Configuration

  • New pytest.ini with test markers (unit, integration, requires_api, etc.)
  • Organized test categorization for better test management
  • VCR cassette markers for recorded API interactions

VCR Test Support

  • tests/cassettes/ infrastructure for recording/replaying API interactions
  • Cost-effective integration testing without repeated API charges
  • Deterministic tests with recorded responses
  • Comprehensive README documenting VCR usage and security

🛠️ Utilities & Infrastructure

Log Filtering System

  • New framework.utils.log_filter module for dynamic log control
  • LoggerFilter class for selective suppression by logger name, level, or pattern
  • Context managers: suppress_logger(), suppress_logger_level(), quiet_logger()
  • Used in health checks to suppress verbose registry initialization logs

Memory Storage Improvements

  • Switched from root logger to framework logger for consistency
  • Reduced log verbosity (INFO → DEBUG for initialization messages)
  • Better integration with framework logging patterns

📚 Documentation Updates

Provider Registry Documentation

  • New ProviderRegistration documentation in registry API reference
  • Custom provider registration guide with Azure OpenAI example
  • Updated component initialization order to include providers
  • Provider access methods documented on RegistryManager

Configuration Documentation

  • Updated provider extensibility information
  • Direct link to custom provider registration guide
  • GitHub issue creation guidance for community providers

🧹 Code Cleanup

Removed Deprecated Code

  • Deleted src/framework/interfaces/openwebui/ (deprecated interface)
  • Removed docs/ressources/other/EXECUTION_POLICY_SYSTEM.md (outdated doc)
  • Cleaned up unused provider logic from multiple files

Test Updates

  • Fixed config import paths (moved from configs.config to framework.utils.config)
  • Added minimal config.yml in tests to prevent config loading errors
  • Updated integration tests for new architecture

Benefits Summary

  1. Maintainability: Single source of truth for provider logic
  2. Extensibility: Applications can register custom providers easily
  3. Consistency: Same adapter used by factory.py and completion.py
  4. Type Safety: Strong typing throughout provider system
  5. Discoverability: registry.list_providers() shows available providers
  6. Clean Architecture: Clear separation between provider interface and implementation
  7. Reduced Code: ~860 lines removed, ~1,090 lines of well-structured adapter code added
  8. Zero Breaking Changes: All existing APIs work identically
  9. Better Testing: VCR infrastructure for integration tests
  10. Enhanced Health Checks: Provider-specific health validation through registry

Technical Details

Component Architecture

Registry System
├── ProviderRegistration (minimal metadata: module_path, class_name)
├── Provider Adapters (metadata as class attributes)
│   ├── BaseProvider (abstract interface)
│   ├── AnthropicProviderAdapter
│   ├── OpenAIProviderAdapter
│   ├── GoogleProviderAdapter
│   ├── OllamaProviderAdapter
│   └── CBorgProviderAdapter
└── RegistryManager (lazy loading, metadata introspection)

Initialization Order

  1. Context classes
  2. Data sources
  3. Providers (new in v0.7.6)
  4. Core nodes
  5. Services
  6. Capabilities
  7. Framework prompt providers

Provider Interface

Each provider implements:

  • create_model() - PydanticAI model creation
  • execute_completion() - Direct API calls
  • check_health() - Connectivity validation

Migration Notes

No breaking changes. The provider registry is a backward-compatible internal refactoring. All existing code using get_model() and get_chat_completion() continues to work without modifications.

For Advanced Users

If you need custom AI providers, you can now register them through the registry system. See documentation for complete examples.

Configuration Changes

No configuration changes required. All existing config.yml provider configurations remain valid.

Files Changed

  • 12 atomic commits implementing provider registry system
  • 3 documentation files updated
  • 6 version files updated for v0.7.6 release

Upgrade Instructions

pip install --upgrade alpha-berkeley-framework

Or with specific extras:

pip install --upgrade "alpha-berkeley-framework[memory,docs]"

Full Changelog: See CHANGELOG.md for detailed commit history

Assets 2
Loading