Modules
Phoenix-native LLM evaluation workbench.
Context for reusable evaluation datasets and suite population.
A reusable collection of ordered evaluation examples.
An ordered single-turn or multi-turn example in a dataset.
Converts documents between formats for LLM consumption.
Context for managing evaluation suites, test cases, quality policies, and runs.
Backward-compatible facade for behaviour-driven evaluation metrics.
Parses and validates assertion payloads from suite editor forms.
Validates and persists uploaded documents for suite test cases.
Loads and executes versioned JSON or YAML suite manifests.
Versioned rubric templates for common model-backed evaluation tasks.
Contract and shared helpers for evaluation metrics.
Evidence available to an evaluation metric.
Normalized execution details for the evaluator behind a metric result.
Registry for the built-in evaluation metrics.
Normalized result returned by every evaluation metric.
Evaluates generated output against a user-defined rubric with an Aludel provider.
Validates and evaluates versioned suite quality policies.
A normalized, versioned representation of an evaluation suite run.
Renders evaluation reports for people and CI systems.
Renders a concise, plain-text evaluation summary.
Renders GitHub Actions workflow-command annotations.
Renders the versioned evaluation report as JSON.
Renders evaluation results as JUnit XML for CI test-report viewers.
Configuration and deterministic reduction for repeated test case execution.
Schema for evaluation test suites.
Immutable version of a quality policy attached to an evaluation suite.
Schema for tracking suite execution results.
Owns suite execution launch under explicit supervision.
Schema for individual test cases within a suite.
Schema for file attachments associated with test cases.
Coordinates suite test case editing workflows outside the web layer.
Imports suite test cases from CSV and JSON payloads.
Adds evaluation assertions to ExUnit tests.
Shared execution boundary for native provider calls and host-app callbacks.
Behaviour and configuration helpers for execution mode dispatch.
Validates file content against claimed MIME types using magic bytes.
Generic HTTP client adapter behaviour.
Converts the first page of a PDF to PNG with bounded ImageMagick resources.
Behaviour for document conversion adapters.
Default LLM HTTP client using ReqLLM.
Behaviour for LLM provider implementations.
Configuration utilities for LLM provider implementations.
Shared error parsing logic for LLM provider adapters.
Anthropic Claude API adapter implementation.
Google Gemini LLM provider implementation.
Groq LLM provider implementation through ReqLLM.
Ollama API adapter implementation.
OpenAI LLM provider implementation.
OpenRouter LLM provider implementation through ReqLLM.
xAI LLM provider implementation through ReqLLM.
AWS S3-backed document storage adapter.
Behaviour for the AWS storage adapter's S3 client boundary.
Default AWS client implementation backed by ExAws.S3.
ExAws HTTP client implementation backed by Req.
Google Cloud Storage-backed document storage adapter.
Behaviour for the GCS storage adapter's Google API client boundary.
Default GCS client implementation backed by google_api_storage and Goth.
Local filesystem adapter for document storage.
Behaviour for external document storage adapters.
LLM client abstraction for multi-provider support.
Context for managing projects that organize prompts and test suites.
Schema for organizing prompts and test suites into projects.
Context for managing prompts and their versions.
Functions for analyzing prompt version evolution and performance metrics.
Export serialization for prompt evolution metrics.
Multi-objective prompt analysis and failure-grounded prompt suggestions.
Schema for managing prompts.
A failure-grounded prompt revision awaiting an explicit human decision.
Schema for prompt versions.
Context for managing AI provider configurations.
Resolves pricing for a given provider and model.
Schema for AI provider configurations.
Curated and generated adversarial evaluation cases.
Stable, versioned adversarial cases used by Aludel.RedTeam.
A validated, reviewable adversarial case proposed by a generator model.
The review boundary returned by Aludel.RedTeam.generate/2.
Central repository accessor for Aludel.
Context for managing runs and run results.
Result of executing a run across one or more providers.
Owns run launch and provider execution under explicit supervision.
Schema for managing runs.
Schema for run results.
Recent activity feeds and time-series activity reporting.
Cost reporting and breakdowns for the dashboard.
Latency reporting and provider breakdowns.
Top-line dashboard metrics and bounded period comparisons.
Domain-facing facade for document storage backends.
Resolves storage configuration without coupling adapters to the storage facade.
Serves pre-compiled static assets for Aludel dashboard.
Loads dashboard session context and enforces LiveView access decisions.
Provides core UI components.
LiveView hooks for common functionality across all LiveViews.
Behavior for customizing dashboard access and user identification.
Provides the aludel_dashboard macro for mounting Aludel in host apps.
Mix Tasks
Runs one Aludel evaluation suite and emits a report.
Installs Aludel by copying migrations to the host application.
Runs Aludel seed data.