warning

Legacy managed-runtime reference. Command examples below describe the older desktop/managed llama.cpp preview and are legacy, not the current BCCLI v0.4 command surface. Use BCCLI and Local AI for current instructions.

Choose a model

Goal

Pick a model from the catalog that can do your task and fits your machine.

What you need

Your task (chat, coding, vision, and so on).
Your machine's RAM and GPU, from botconnector doctor (Install the Local Runtime) or System requirements.

Steps

Name your task. The capability you need determines the candidate set: chat, coding, reasoning, vision, embeddings, audio — see Capabilities.
Open the Model catalog and filter by that capability. Every entry carries evidence labels inferred from Hub metadata — they are hints, not guarantees.
Filter by hardware fit. The catalog highlights models whose file sizes fit your RAM — see Hardware fit.
Compare candidates on quality vs size. Larger parameter counts and larger quantizations give better answers but need more memory — see Quantization and Context and memory.
Choose a model and download it: Download a model for device runs, or pick it in the Model picker for cloud runs.

Expected result

You have one model picked, with a quantization that fits your machine and a compute target you chose yourself.

If it fails

The chosen model does not behave as labelled (wrong template, weak output): Model problems.
Model does not fit your memory: Memory problems and Hardware fit.
General problems: Troubleshooting.

Next step

Match the model to your machine in detail: Hardware fit, or go straight to downloading: Download a model.