Modules
Fast, persistent_term-backed LLM model metadata catalog.
OTP Application for LLMDB.
Configuration reading and normalization for LLMDB.
Shim to call DeepMerge.deep_merge/3 with a 3-arity resolver without Dialyzer false positives.
Pure ETL pipeline for BUILD-TIME LLM model catalog generation.
Lightweight, deterministic enrichment of model data.
Auto-generated module containing all valid modality atoms.
Auto-generated module containing all valid provider atoms.
Read-only runtime access to generated model history artifacts.
Backfills model history by diffing committed provider snapshots across git history.
Handles loading and merging of packaged snapshots with runtime customization.
Precedence-aware merging with exclude handling for LLM model data.
Model struct with Zoi schema validation.
Complete normalization utilities for raw data into consistent formats.
Provides access to the packaged base snapshot.
Pricing pipeline for converting legacy cost data and applying provider defaults.
Provider struct with Zoi schema validation.
Query functions for selecting models based on capabilities and requirements.
Runtime configuration compilation for consumer applications.
Unified data source interface for LLMDB.
Remote source for Anthropic models (https://api.anthropic.com/v1/models).
Remote source for Google Gemini models (https://generativelanguage.googleapis.com/v1beta/models).
Loads model metadata from local TOML files in a directory structure.
Remote source for models.dev metadata (https://models.dev/api.json).
Remote source for OpenAI models (https://api.openai.com/v1/models).
Remote source for OpenRouter metadata (https://openrouter.ai/api/v1/models).
Remote source for xAI (Grok) models (https://api.x.ai/v1/models).
Remote source for Zenmux models (https://zenmux.ai/api/v1/models).
Canonical "provider:model" spec parsing and resolution.
Manages persistent_term storage for LLM model snapshots with atomic swaps.
Validation functions for providers and models using Zoi schemas.
Mix Tasks
Builds snapshot.json from configured sources using the Engine ETL pipeline.
Backfills model history from committed provider snapshots.
Checks whether priv/llm_db/history is current with provider metadata history.
Incrementally syncs model history from committed provider snapshots.
Install and configure LLM DB for use in an application.
Lists all models from the LLMDB catalog with lifecycle status and aliases.
Pulls latest model metadata from all configured remote sources and caches locally.
Updates the version in mix.exs to CalVer format (YYYY.M.PATCH).