Skip to content

CHANGELOG.md

Latest commit

 

History

History
198 lines (165 loc) · 8.65 KB

CHANGELOG.md

File metadata and controls

198 lines (165 loc) · 8.65 KB

Changelog

All notable changes to the TealTiger Python SDK will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased

1.1.1 - 2026-04-03

Fixed

  • README rewritten to accurately reflect all features included in v1.1.0
  • Removed misleading "Enterprise Feature Comparison" table that incorrectly suggested v1.1.0 had fewer features than a non-existent "v1.1.x Enterprise"
  • Removed "What's New in v0.2.0" section
  • Removed merge conflict markers
  • Removed old URLs (tealtiger.co.in)

Changed

  • License updated from MIT to Apache 2.0 across all files (LICENSE, pyproject.toml classifier, README badges)
  • All enterprise features (TealEngine, TealGuard, TealCircuit, TealAudit, correlation IDs, policy testing) now correctly presented as included v1.1.0 features

Notes

  • No code changes — documentation and metadata only
  • Fully backward compatible with v1.1.0

1.1.0 - 2026-03-15

Added

  • TealEngine — Deterministic policy evaluation with multi-mode enforcement (ENFORCE, MONITOR, REPORT_ONLY)
  • TealGuard — Client-side security guardrails (PII detection, prompt injection, content moderation)
  • TealCircuit — Circuit breaker for cascading failure prevention
  • TealAudit — Versioned audit logging with security-by-default PII redaction
  • Correlation IDs — Auto-generated UUID v4 with OpenTelemetry-compatible trace IDs
  • Decision Contract — Deterministic typed Decision object with risk scores and reason codes
  • Policy Test Harness — CLI/library test runner with JUnit XML export for CI/CD
  • Multi-Provider Support — 7 providers (OpenAI, Anthropic, Gemini, Bedrock, Azure OpenAI, Cohere, Mistral)
  • OWASP Coverage — 7/10 ASIs covered with SDK-only architecture

0.2.2 - 2026-01-31

Added

  • Cost Tracking & Budget Management - Complete feature parity with TypeScript SDK v0.2.2

    • CostTracker - Track AI model costs across OpenAI, Anthropic, and Azure OpenAI
    • BudgetManager - Create and enforce budgets with alerts and blocking
    • InMemoryCostStorage - Store and query cost records
    • Support for 20+ AI models with accurate pricing
    • Custom pricing support for proprietary models
    • Budget periods: hourly, daily, weekly, monthly, total
    • Alert thresholds with severity levels (info, warning, critical)
    • Agent-scoped budgets for multi-agent systems

  • Guarded AI Clients - Drop-in replacements with integrated security

    • GuardedOpenAI - Secure OpenAI client with guardrails and cost tracking
    • GuardedAnthropic - Secure Anthropic client with guardrails and cost tracking
    • GuardedAzureOpenAI - Secure Azure OpenAI client with deployment mapping
    • Automatic input/output guardrail execution
    • Pre-request budget checking and enforcement
    • Automatic cost calculation and recording
    • Security metadata in all responses

  • Example Scripts - Comprehensive demos for all new features

    • cost_tracking_demo.py - Cost estimation and tracking examples
    • budget_management_demo.py - Budget creation and enforcement examples
    • guarded_openai_demo.py - GuardedOpenAI usage examples
    • guarded_anthropic_demo.py - GuardedAnthropic usage examples
    • guarded_azure_openai_demo.py - GuardedAzureOpenAI usage examples

Features

  • Multi-Provider Support: OpenAI, Anthropic, Azure OpenAI
  • Accurate Pricing: Real-time cost calculation for 20+ models
  • Budget Enforcement: Block requests that exceed budgets
  • Alert System: Configurable thresholds with severity levels
  • Agent Isolation: Separate budgets per agent
  • Cost Queries: Query costs by agent, date range, request ID
  • Custom Pricing: Override pricing for custom models
  • Deployment Mapping: Azure deployment names to model names
  • Security Integration: Guardrails + cost tracking in one client

Performance

  • Async-first design for all operations
  • Efficient in-memory storage with O(1) lookups
  • Parallel guardrail execution
  • < 10ms cost calculation overhead

Documentation

  • Updated README with cost tracking and guarded clients sections
  • Added 5 comprehensive example scripts
  • Full API documentation for all new classes
  • Migration guide from v0.2.0

Dependencies

  • Added openai>=1.0.0 for GuardedOpenAI and GuardedAzureOpenAI
  • Added anthropic>=0.18.0 for GuardedAnthropic
  • Added hypothesis>=6.0.0 for property-based testing (dev)

Testing

  • 71+ new tests for cost tracking and guarded clients
  • Property-based tests for correctness validation
  • Integration tests for end-to-end workflows
  • 61% overall test coverage (focused on new features)

Notes

  • Feature Parity: Python SDK now matches TypeScript SDK v0.2.2
  • Breaking Changes: None - fully backward compatible
  • Migration: Existing code continues to work without changes

0.2.0 - 2026-01-30

Added

  • Client-Side Guardrails - Offline security protection without server dependency
    • GuardrailEngine for parallel/sequential guardrail execution
    • PIIDetectionGuardrail - Detect and redact PII (emails, phones, SSNs, credit cards)
    • ContentModerationGuardrail - Detect harmful content (hate, violence, harassment)
    • PromptInjectionGuardrail - Detect jailbreak and injection attempts
    • Configurable actions: block, allow, redact, mask, transform
    • Timeout protection and error handling with asyncio
    • Pydantic models for type safety
  • Comprehensive test suite for guardrails (50 tests passing)
  • Guardrails demo example with real-world scenarios
  • Full async/await support for all guardrail operations

Features

  • Offline Capability: Run guardrails without network calls
  • Parallel Execution: Execute multiple guardrails simultaneously with asyncio
  • Flexible Actions: Block, redact, mask, or transform risky content
  • Risk Scoring: Quantify security risks (0-100 scale)
  • Pattern Detection: Regex-based detection with high accuracy
  • OpenAI Integration: Optional OpenAI Moderation API support
  • Type Safety: Full Pydantic models for all guardrail results

Performance

  • < 50ms guardrail execution (parallel mode)
  • Configurable timeouts per guardrail
  • Efficient pattern matching with compiled regex
  • Async-first design for high concurrency

Documentation

  • Added guardrails usage examples
  • Updated README with guardrails showcase
  • Added inline documentation for all guardrail classes

0.1.1 - 2026-01-29

Fixed

  • Package name changed to agentguard-sdk (from agentguard) due to PyPI name conflict
  • Updated all imports and documentation

Added

0.1.0 - 2026-01-28

Added

  • Initial release of AgentGuard Python SDK
  • Core security evaluation functionality
  • Tool execution with security decisions (allow/deny/transform)
  • Security Sidecar Agent (SSA) HTTP client
  • Configuration management with validation
  • Comprehensive error handling with custom exceptions
  • Audit trail functionality
  • Policy validation and management
  • Full async/await support
  • Type hints throughout the codebase
  • Comprehensive test suite with pytest
  • Examples for basic and advanced usage
  • Complete API documentation

Features

  • Security Evaluation: Evaluate tool calls before execution
  • Policy Enforcement: Automatic policy-based decision making
  • Request Transformation: Safe transformation of risky operations
  • Audit Trail: Complete audit logging for compliance
  • Performance: < 100ms security evaluation overhead
  • Type Safety: Full type hints with Pydantic models
  • Async Support: Built-in async/await for modern Python

Security

  • API key authentication with SSA
  • Input validation and sanitization
  • Secure HTTP communication with httpx
  • Error handling that doesn't leak sensitive information

Developer Experience

  • Comprehensive documentation with examples
  • Type hints for better IDE support
  • Pytest test suite with 100% core functionality coverage
  • Examples for common integration patterns
  • Poetry and pip support