Skip to content

CHANGELOG.md

Latest commit

 

History

History
94 lines (76 loc) · 3.39 KB

CHANGELOG.md

File metadata and controls

94 lines (76 loc) · 3.39 KB

Changelog

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

[1.0.0] - 2025-12-27

Added

Core Features - The Four Pillars

Prompt Composer

  • Tokenizer service using tiktoken-php with support for 46+ models
  • Section management with priorities for intelligent prompt assembly
  • Shrinkers (Truncate, Compress) for fitting content to context windows
  • Mindwave::prompt() facade for fluent prompt building
  • Automatic token counting and context window management

OpenTelemetry Tracing

  • Full GenAI semantic conventions implementation
  • Database exporter for local development and queries
  • OTLP exporter for production observability (Jaeger, Grafana, Datadog, Honeycomb)
  • Multi-exporter support for fan-out to multiple backends
  • Automatic cost estimation and token usage tracking
  • Events system (RequestStarted, TokenStreamed, ResponseCompleted, ErrorOccurred)
  • Artisan commands: mindwave:trace-export, mindwave:trace-prune, mindwave:trace-stats

Streaming SSE

  • streamText() method on LLM interface
  • streamChat() method for structured streaming with metadata
  • StreamChunk DTO for consistent streaming responses
  • StreamingException with retryable flag and provider tracking
  • StreamRetryHandler with exponential backoff and jitter
  • SSE formatter (StreamedTextResponse) for EventSource compatibility
  • Client examples: Vanilla JS, Alpine.js, Vue.js, React, TypeScript

TNTSearch Context Discovery

  • TntSearchSource for Eloquent, arrays, and CSV data
  • VectorStoreSource for Brain/RAG integration
  • EloquentSource for SQL LIKE searches
  • StaticSource for hardcoded context
  • ContextPipeline for multi-source aggregation and deduplication
  • EphemeralIndexManager for temporary search indexes with TTL
  • Artisan commands: mindwave:index-stats, mindwave:clear-indexes

LLM Drivers

  • OpenAI - Full support including GPT-4, GPT-4 Turbo, GPT-4o, o1 models
  • Anthropic - Claude 3 Opus, Sonnet, Haiku; Claude 3.5 Sonnet
  • Mistral - Mistral Medium, Large, and other models
  • Fake - Testing driver with configurable responses

Vector Store Drivers

  • Pinecone - Cloud vector database
  • Qdrant - Open-source vector database
  • Weaviate - AI-native vector database
  • InMemory - For testing and small datasets
  • File - JSON file-based persistence

Document Loaders

  • PDF loader (using smalot/pdfparser)
  • Word document loader (.docx, .odt)
  • HTML loader with tag stripping
  • URL loader with content extraction
  • Plain text loader

Integrations

  • Laravel Telescope - MindwaveWatcher for request monitoring
    • Tags: mindwave, provider:, model:, slow, expensive, cached
    • Privacy controls with capture_messages option

Changed

  • Namespace updated to Mindwave\Mindwave
  • Removed agent/crew orchestration code (pivot to utilities focus)
  • LLM interface now requires streamText() and streamChat() methods

Fixed

  • Mistral driver config key resolution
  • Weaviate dependency installation
  • Model class renamed to ModelNames to avoid conflicts
  • Tool choice API compatibility across providers

Developer Experience

  • 1300+ tests with Pest PHP
  • PHPStan level 4 with baseline
  • Laravel Pint code style
  • GrumPHP pre-commit hooks
  • GitHub issue/PR templates
  • Dependabot for dependency updates
  • CONTRIBUTING.md and SECURITY.md

[0.x] - Pre-release

Initial development versions with agent framework (deprecated).