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

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

Unreleased

Planned

  • AWS Bedrock provider
  • Parallel tool execution
  • Audio input/output support
  • Token counting utilities
  • Cost estimation

0.3.0 - 2026-01-21

Added

New Provider

  • OpenRouter - Access 100+ models via unified API with openrouter/ prefix

Testing Infrastructure

  • ExVCR integration - HTTP recording for deterministic tests without API keys
  • VCR tests - 10 cassette-based tests for OpenAI, Anthropic, and Gemini
  • API key filtering - Automatic removal of sensitive data from cassettes

Gemini Models

  • Gemini 2.5 series - gemini-2.5-pro-preview-05-06, gemini-2.5-flash-preview-05-20
  • Gemini 2.0 nano - gemini-2.0-flash-lite for lowest latency

Changed

  • Provider architecture - Extracted shared logic into ElixirLLM.Providers.Base
  • Streaming implementation - Refactored for Req 0.5.x compatibility using process dictionary pattern
  • Code deduplication - Reduced provider code by ~40% through shared helpers

Fixed

  • Streaming bug - Fixed {:fold, acc, fun} format not supported in Req 0.5.x
  • Dialyzer warnings - Added :mix to PLT for Mix.Task analysis

0.2.0 - 2026-01-20

Added

New Providers

  • Google Gemini - gemini-2.0-flash, gemini-1.5-pro, gemini-1.5-flash support
  • Mistral AI - Direct API support for mistral-large, codestral, etc.
  • Groq - Ultra-fast LPU inference with llama-3.3-70b-versatile, llama4-scout
  • Together AI - 100+ open models via together/ prefix

Resilience Features

Error Handling

  • Structured errors - 9 custom exception types for different failure modes
  • Error classification - retryable?/1 to identify transient vs permanent failures
  • Error conversion - from_response/2 to parse provider errors into structured types
  • Error types: APIError, RateLimitError, AuthenticationError, ValidationError, NetworkError, TimeoutError, ProviderError, ToolError, MaxDepthError

Changed

  • Provider detection - Now supports 8 providers with automatic model-based routing
  • Test coverage - Expanded from 24 to 60 tests

Fixed

  • Code complexity - Replaced nested case in media_type_from_path/1 with efficient Map lookup
  • List operations - Fixed inefficient length(list) > 0 checks in all providers
  • Enum operations - Used Enum.map_join/3 instead of Enum.map |> Enum.join
  • Alias ordering - All modules now have alphabetically sorted aliases

0.1.0 - 2026-01-20

Added

Core Features

  • Unified Chat API - Single interface for all LLM providers
  • Pipe-based configuration - Idiomatic Elixir with new() |> model() |> ask()
  • Conversation continuity - Multi-turn conversations with automatic context
  • Streaming support - Real-time token streaming with callbacks

Providers

  • OpenAI - GPT-4o, GPT-4-turbo, o1, o3-mini support
  • Anthropic - Claude Sonnet 4, Claude Opus 4 support
  • Ollama - Local models (Llama, Mistral, CodeLlama)
  • Auto-detection - Provider automatically detected from model name

Tools

  • Tool behaviour - use ElixirLLM.Tool with DSL for parameters
  • Inline tools - ElixirLLM.Tool.define/1 for quick definitions
  • Automatic execution loop - No manual tool call handling required
  • Tool callbacks - on_tool_call and on_tool_result hooks

Structured Output

  • Schema DSL - use ElixirLLM.Schema with field, embeds_one, embeds_many
  • JSON Schema generation - Automatic conversion for provider APIs
  • Type parsing - Response automatically parsed into Elixir structs

Embeddings

  • Vector generation - ElixirLLM.embed/2 for text embeddings
  • Batch support - Embed multiple texts in one call
  • Similarity functions - cosine_similarity/2, euclidean_distance/2, dot_product/2
  • OpenAI text-embedding-3-small - Default embedding model

Multi-Modal

Phoenix/Ecto Integration

  • Mix generator - mix elixir_llm.gen.ecto creates schemas and migrations
  • Chat persistence - Automatic message saving with use ElixirLLM.Ecto.Chat
  • Message schema - Token tracking, role enum, tool call references
  • Tool call schema - Full tool execution history

Observability

  • Telemetry integration - Events for chat, streaming, tools, embeddings
  • Request/response metadata - Model, tokens, duration tracking

Technical Details

  • Minimal dependencies: Req, Jason, Telemetry, NimbleOptions
  • Elixir 1.14+ compatibility
  • Full documentation with examples

0.0.1 - 2026-01-20

Added

  • Initial project structure
  • Basic README