Local Model Registry & Scanner

FineTuneMyAI automatically discovers on-device LLMs installed across your machine without transmitting full filesystem paths to the cloud control plane.

Supported Discovery Sources

1. Ollama Daemon

Inspects ~/.ollama/models or connects to the local Ollama daemon at 127.0.0.1:11434.

2. Hugging Face Hub

Scans standard snapshot trees in ~/.cache/huggingface/hub for safetensors weights.

3. Apple MLX & Local Caches

Discovers Apple Silicon MLX models in ~/.cache/mlx and user model directories without touching server storage.

Scanning & Inspecting via Native Agent CLI

End users can discover, list, and inspect resident models directly through the native standalone agent CLI without needing Git, npm, or repo source code:

# 1. Scan resident models and sync metadata to your dashboard
$ finetunemyai models scan

# 2. View table of resident models on this machine
$ finetunemyai models list

# 3. Inspect architectural specs, context length, and memory readiness
$ finetunemyai models inspect qwen2.5-coder:7b-instruct

[+] Discovered 2 local models:
• qwen2.5-coder:7b-instruct (7B parameters • 4-bit Ollama GGUF • VALID)
• Qwen2.5-0.5B-Instruct-4bit (0.5B parameters • 4-bit Apple MLX • VALID)

Opaque Model Identifiers

The local agent computes a deterministic SHA-256 hash from the base model's canonical name, creating an opaque identifier like model_qwen25_7b. The central database stores only this opaque string and model metadata (parameter count, architecture family, quantization format). Your physical path remains local to your device.

← Device Pairing Protocol Corpus Quality & Splits →