Aludel 0.7.0 is released

Since 0.5.0, Aludel has grown into a much more complete LLM evaluation toolkit for Elixir and Phoenix applications.

Highlights

• Rubric-based LLM judges with seven built-in, versioned templates
• Repeated evaluation sampling with all, any, majority, and minimum pass-rate strategies
• Quality policies covering pass rate, evaluator score, cost, latency, and metadata
• Versioned evaluation suites defined in JSON or YAML
• ExUnit helpers for running LLM evaluations alongside application tests
• Console, JSON, JUnit, and GitHub-compatible reports for local development and CI
• Curated and generated red-team datasets with review, provenance, and idempotent imports
• First-class xAI, Groq, and OpenRouter support, alongside OpenAI, Anthropic, Gemini, and Ollama
• A richer LiveView dashboard for configuring judges, sampling, policies, reports, and red-team cases
• Local, AWS S3, and Google Cloud Storage support in the standalone distribution

The full evaluation loop

The goal is to cover the entire LLM evaluation workflow:

Develop prompts → compare models and versions → catch regressions → inspect quality, cost, and latency trade-offs → test adversarial cases → enforce quality thresholds in CI

Aludel can be used through its LiveView UI, Mix tasks, ExUnit integration, or public Elixir APIs.

• GitHub: GitHub - ccarvalho-eng/aludel: Phoenix-native LLM evaluation, prompt testing, model comparison, red teaming, and observability for Elixir · GitHub

• Hex: aludel | Hex

• Documentation: aludel v0.7.0 — Documentation

• Wiki & examples: Home · ccarvalho-eng/aludel Wiki · GitHub

Screenshot 2026-09-05 at 9.00.36 AM1200×777 25.6 KB

Screenshot 2026-09-05 at 9.00.46 AM1200×773 20.1 KB

Screenshot 2026-09-05 at 9.01.03 AM1200×757 22.6 KB

Screenshot 2026-09-05 at 9.01.17 AM1200×792 37.9 KB