llamero

Run AI models locally on Apple Silicon from Crystal with the llamero shard (Llamero::Native, MLX/Metal). Use when the user wants on-device or local LLM inference, local chat, token streaming, structured JSON output from a local model, or to load/keep a model resident in memory without cloud APIs like OpenAI. Also covers downloading mlx-community models and activating LoRA adapters.

Train LoRA/QLoRA adapter "filters" on a local model with the llamero shard (Crystal, Apple Silicon). Use when the user wants to fine-tune a local model, teach a model new facts or documents (manuals, internal docs, domain knowledge), create/train a LoRA or QLoRA adapter, build a golden dataset, or toggle learned knowledge on and off without reloading the model. No Python required.

Call cloud AI APIs (OpenAI, Anthropic/Claude, Groq, OpenRouter) from Crystal with the llamero shard - chat, streaming, structured JSON output, embeddings, and automatic failover between providers. Use when the user wants to call an LLM API from Crystal, needs provider failover/retries, or wants typed structured responses from a cloud model.