LLM DB - LLM Model Metadata Package

This package was extracted out of the ReqLLM project.

LLM DB is a model metadata catalog with fast, capability-aware lookups. Use simple "provider:model" or "model@provider" specs, get validated Provider/Model structs, and select models by capabilities. Ships with a packaged snapshot; no network required by default.

• Primary interface: model_spec — a string like "openai:gpt-4o-mini" or "gpt-4o-mini@openai" (filename-safe)
• Fast O(1) reads via :persistent_term
• Minimal dependencies

Why?

When building ReqLLM, we implemented a model metadata system by pulling from https://models.dev. This worked well initially, but became challenging as we discovered various issues with LLM APIs. We submitted many PR’s upstream to models.dev, but they built their database for their purposes and it became obvious that our needs were diverging.

This package was extracted because it will have automated releases weekly to capture the latest model releases as quickly as possible.

It also standardizes the “model spec” - a unique string that can be used to address a specific model + provider combo. We support various spec formats.

For consumers, this package also supports filtering, local model definitions and a really nice allow/deny system so even when we have 1200 models in our database, but your app only wants to support 5, you can easily configure this.

Hex Release: llm_db | Hex
Github:

This package is part of the Agent Jido ecosystem.