This Elixir library, instructor_ex, enables structured data output from Large Language Models (LLMs) by leveraging Ecto schemas. It targets Elixir developers seeking to integrate LLM capabilities into their applications, providing a robust way to ensure LLM responses are not only correctly formatted but also semantically valid through Ecto's validation system.

How It Works

The library acts as a spiritual port of the Python Instructor library, abstracting the complexity of JSON schema generation. Instead, it uses Elixir's Ecto schemas, including @llm_doc attributes for LLM descriptions and validate_changeset/1 functions for validation rules. The use Instructor macro integrates these schemas, allowing Instructor.chat_completion/1 to request structured responses directly mapped to Ecto schemas. The max_retries parameter facilitates automatic iterative correction of LLM output based on validation failures.

Quick Start & Requirements

• Install via Mix.install([:instructor]) or add {:instructor, "~> 0.1.0"} to mix.exs.
• Requires an LLM API key (e.g., OpenAI) and potentially access to specific models.
• Official Quickstart Guide: https://github.com/thmsmlr/instructor_ex#quickstart-guide
• Elixir Denver Meetup Talk: https://www.youtube.com/watch?v=RABXu7zqnT0

Highlighted Details

• Supports OpenAI, Anthropic, Groq, Ollama, Gemini, vLLM, and llama.cpp.
• Leverages Ecto schemas and validations for LLM output structuring and validation.
• Automatic retries for correcting LLM output based on Ecto changeset validations.
• Eliminates the need for manual JSON schema definition.

Maintenance & Community

• Project appears to be maintained by a single author, thmsmlr.
• No explicit community links (Discord/Slack) or roadmap are provided in the README.

Licensing & Compatibility

• The README does not explicitly state a license. Based on typical open-source Elixir projects, it is likely MIT or Apache 2.0, but this requires verification.
• Compatible with Elixir applications. Commercial use depends on the underlying LLM provider's terms and the project's license.

Limitations & Caveats

The library is in its early stages (v0.1.0), indicating potential for breaking changes and missing features. The lack of explicit licensing information could be a concern for commercial adoption.