Skip to content

CHANGELOG.md

Latest commit

 

History

History
353 lines (254 loc) · 11.2 KB

CHANGELOG.md

File metadata and controls

353 lines (254 loc) · 11.2 KB

Changelog

[Unreleased] - Unreleased

[1.5.0] - 2025-09-30

Added

  • CMake option LLMCPP_USE_OPENSSL to optionally use OpenSSL for HTTPS (defaults to OFF)
  • Native SSL support via platform-specific implementations:
    • macOS: SecureTransport (via Security framework)
    • Windows: WinSSL/Schannel (via crypt32)
    • Linux: Can enable OpenSSL with -DLLMCPP_USE_OPENSSL=ON

Changed

  • Default SSL/TLS implementation changed from OpenSSL to native platform SSL
  • OpenSSL is now an optional dependency instead of required
  • Faster builds and smaller binaries due to removed OpenSSL dependency

Fixed

  • Simplified dependency management by using native SSL implementations

[1.4.0] - 2025-09-30

Added

  • MCP (Model Context Protocol) tools integration with OpenAI Responses API
  • OpenAI::McpTool struct for configuring MCP servers
  • OpenAIMcpUtils namespace with helper functions:
    • extractMcpCalls() - Extract all MCP tool calls from response
    • wasToolCalled() - Check if specific tool was called
    • getToolOutput() - Get output from specific tool
    • getAllToolOutputs() - Get all tool outputs
    • wereMcpToolsListed() - Check if tools were listed
    • getAvailableMcpTools() - Get list of available tools
    • wereAllToolsCalled() - Check if all expected tools were called
    • getMcpUsageStats() - Get usage statistics
  • Helper functions: setTools(), hasTools(), getToolsJson()
  • Integration tests for MCP functionality using public DeepWiki MCP server
  • Generic extensions field in LLMRequestConfig for provider-specific data
  • JsonUtils.h with safe JSON parsing utilities

Changed

  • Refactored core LLM types to be provider-agnostic
  • Updated to support gpt-5-mini model

Other

  • Merge pull request #60 from lucaromagnoli/feature/mcp-tools-integration

[1.3.0] - 2025-08-11

Other

  • Merge pull request #57 from lucaromagnoli/release/v1.2.0
  • Include ResponseParser.h in main llmcpp.h header
  • Merge pull request #58 from lucaromagnoli/fix/expose-response-parser

[1.2.0] - 2025-08-10

Added

  • Add ResponseParser for provider-agnostic structured response parsing
  • implement comprehensive LLM response parsing
  • enhance ResponseParser to handle direct function tags

Other

  • Merge pull request #55 from lucaromagnoli/release/v1.1.1
  • security: remove anthropic_real_response.json
  • Fix compilation issue in ResponseParser.cpp
  • Implement parseDirectFunctionTags with simplified string-based approach
  • Fix parseDirectFunctionTags to be generic and handle unclosed tags
  • Fix parseDirectFunctionTags to throw exception for missing functionName
  • Fix isAnthropicResponse to detect any XML-like tags
  • Fix formatting after pre-commit hooks
  • SECURITY: Remove dangerous test files that leaked aideas implementation details
  • Fix formatting after pre-commit hooks
  • Remove unnecessary text wrapping in LLM response conversion
  • Fix Anthropic MessagesResponse parsing to handle tool_use content
  • Update default model in AnthropicConfig to CLAUDE_SONNET_3_7
  • Update default Anthropic model to Claude 4 Sonnet
  • Update AnthropicConfig to use the latest API version
  • Add tool_choice parameter to Anthropic MessagesRequest
  • Replace literal model names with enum usage in getAvailableModels()
  • Update formatting and anthropic version
  • Fix anthropic-version to valid 2023-06-01
  • Fix tool_choice format - should be object not array
  • Fix toLLMResponse with smart parsing parameter
  • Merge pull request #56 from lucaromagnoli/feature/response-parser

[1.1.1] - 2025-08-09

Fixed

  • prevent creation of branch-named tags in release workflow

Other

  • Merge pull request #51 from lucaromagnoli/release/v1.1.0
  • Merge pull request #52 from lucaromagnoli/fix/release-workflow-tag-bug
  • Merge branch 'main' into release/v1.1.0
  • Merge pull request #53 from lucaromagnoli/release/v1.1.0
  • Style: apply pre-commit fixes (clang-format, EOF)
  • Merge pull request #54 from lucaromagnoli/fix/windows-ssl-guards

[1.1.0] - 2025-08-09

Added

  • Complete Anthropic Claude API integration
  • Add unified OpenAI vs Anthropic benchmark tests

Fixed

  • Anthropic API integration bug fixes
  • resolve macOS CI build issues

Documentation

  • Add comprehensive benchmark analysis to README

Testing

  • fix Anthropic unit tests to match corrected message ordering

Build

  • trigger CI workflow
  • remove paths-ignore rule that skips CI on documentation changes
  • refine paths-ignore to only skip docs/ directory
  • skip CI on documentation branches (docs/**)

Maintenance

  • bump version to 1.1.0 and update changelog

Other

  • Merge pull request #50 from lucaromagnoli/feat/anthropic-integration
  • Revert "fix: resolve macOS CI build issues"

[1.1.0] - 2025-08-09

Added

  • Complete Anthropic Claude API integration
  • Add unified OpenAI vs Anthropic benchmark tests

Fixed

  • Anthropic API integration bug fixes

Documentation

  • Add comprehensive benchmark analysis to README

Testing

  • fix Anthropic unit tests to match corrected message ordering

Build

  • trigger CI workflow
  • remove paths-ignore rule that skips CI on documentation changes
  • refine paths-ignore to only skip docs/ directory
  • skip CI on documentation branches (docs/**)

Other

  • Merge pull request #50 from lucaromagnoli/feat/anthropic-integration

[1.0.23] - 2025-08-08

Documentation

  • add GPT-5 mini/nano; usage note for reasoning effort; enums updated

Build

  • add CodeQL, gitleaks, dependency review, and secret-file guard
  • switch to manual gitleaks invocation; set GITHUB_TOKEN; ensure SARIF path exists
  • trigger only on push/schedule; drop PR triggers; make gitleaks non-blocking off main

Maintenance

  • v1.0.23

Other

  • Merge pull request #41 from lucaromagnoli/release/v1.0.22
  • tests(bench): add model-comparison integration benchmark and internal microbenchmarks
  • Merge pull request #45 from lucaromagnoli/ci/security-workflow
  • Merge pull request #46 from lucaromagnoli/docs/readme-gpt5
  • tests(bench): add test_benchmarks.cpp to the test suite and remove obsolete integration benchmark file
  • bench: exclude unsupported models; boost caps for reasoning models; add run_model_benchmarks.sh; CSV output
  • bench: move run_model_benchmarks.sh under tests/bench
  • bench: mark run_model_benchmarks.sh executable
  • bench: avoid max_output_tokens for GPT-5 family; keep caps for others
  • bench: include token usage (input/output/total) in CSV output
  • models: add gpt-5-mini and gpt-5-nano to RESPONSES_MODELS list
  • Merge pull request #47 from lucaromagnoli/feat/model-benchmarks

[1.0.22] - 2025-08-08

Added

  • add GPT-5 model support (enum, mappings, responses list); tests cover model and integration stub

Fixed

  • remove branch restrictions from CI workflow - tests should run everywhere
  • remove pull_request trigger to avoid duplicate expensive builds
  • release workflow should only trigger on merge to main

Other

  • Merge pull request #39 from lucaromagnoli/release/v1.0.21
  • openai: gpt-5 integration hardening — client auto-polls incomplete responses; remove explicit maxTokens in integration tests; drop manual polling in tests; minor debug output tweaks
  • Merge pull request #40 from lucaromagnoli/feat/gpt5

[1.0.21] - 2025-08-07

Fixed

  • Windows compilation error with std::unique_ptr
  • remove pull_request triggers to avoid duplicate CI jobs
  • replace std::unique_ptr with proper httplib::Client
  • add Windows dependencies and configure httplib properly

Other

  • Merge pull request #38 from lucaromagnoli/fix/windows-unique-ptr

[1.0.20] - 2025-08-06

Fixed

  • Fixed Windows MSVC compilation error with std::unique_ptr
  • Replaced std::unique_ptr with raw pointer for SSL placeholder

[1.0.19] - 2025-07-13

Added

  • skip release build-and-test if CI has already passed for the commit

Fixed

  • simplify CI condition to prevent tag builds

Other

  • Merge pull request #30 from lucaromagnoli/feat/add-models
  • Remove o3 model integration test; only test o3-mini (no temperature or maxTokens)
  • Commit remaining changes: core, openai, and unit test updates for optional params and o3-mini logic
  • Fix maxTokens test: now optional instead of defaulting to 200
  • Bump version to 1.0.18 for patch release

[1.0.17] - 2025-07-13

Fixed

  • add explicit condition to skip CI on tag pushes

[1.0.16] - 2025-07-13

Fixed

  • add checkout step to check-open-prs job in release workflow

[1.0.15] - 2025-07-13

Build

  • do not run CI workflow on version tags (only release workflow runs on tags)
  • optimize workflows - merge code quality into CI and skip release on open PRs

[1.0.14] - 2025-07-13

Fixed

  • correct GPT-4.5 model status - it's a current preview model, not deprecated
  • update tests to use correct GPT-4.5 model string (gpt-4.5-preview)
  • implement parameter filtering for reasoning models (O-series)

Documentation

  • update README to clarify model enum usage and remove recommendation references

Maintenance

  • remove duplicate release-workflow.sh script

Other

  • Merge pull request #29 from lucaromagnoli/feat/release-test
  • Remove getRecommendedModel and related tests, clean up OpenAI model enum logic

[1.0.13] - 2025-07-13

Changed

  • remove macOS from release builds for cost optimization

[1.0.12] - 2025-07-13

Added

  • use Python script for robust release notes extraction

Fixed

  • resolve awk syntax error in release notes generation

[1.0.11] - 2025-07-13

Fixed

  • improve release notes generation and changelog handling

Maintenance

  • bump version to 1.0.10 and update changelog

Other

  • Merge pull request #27 from lucaromagnoli/feat/release_process
  • Merge pull request #28 from lucaromagnoli/fix/pipeline-merge

[1.0.10] - 2025-07-13

Other

  • Merge pull request #27 from lucaromagnoli/feat/release_process

[1.0.11] - 2025-07-13

Fixed

  • improve release notes generation and changelog handling

Maintenance

  • bump version to 1.0.10 and update changelog

Other

  • Merge pull request #27 from lucaromagnoli/feat/release_process
  • Merge pull request #28 from lucaromagnoli/fix/pipeline-merge

[1.0.9] - 2025-07-13

Added

  • add interactive release workflow with changelog editing

Other

  • Merge pull request #26 from lucaromagnoli/feat/cost-opt

[1.0.11] - 2025-07-13

Fixed

  • improve release notes generation and changelog handling

Maintenance

  • bump version to 1.0.10 and update changelog

Other

  • Merge pull request #27 from lucaromagnoli/feat/release_process
  • Merge pull request #28 from lucaromagnoli/fix/pipeline-merge

[1.0.8] - 2025-07-13

Added

  • optimize GitHub Actions costs by removing macOS from regular CI
  • streamline release process to eliminate double PR workflow

Other

  • Merge pull request #23 from lucaromagnoli/improve-async
  • Merge branch 'main' into release/v1.0.7
  • Merge pull request #24 from lucaromagnoli/release/v1.0.7
  • Merge pull request #25 from lucaromagnoli/cost-optimise-pipeline

All notable changes to this project will be documented in this file.

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

[1.0.8] - 2025-07-13

Added

  • add type-safe Model enum support with convenience methods
  • modernize CMake for proper install and export

Other

  • Merge pull request #17 from lucaromagnoli/fix/llmrequest