Catalogo strumenti

Il software che ti serve per eseguire davvero un modello AI.

Un modello è solo un file. Per usarlo serve un software che carichi quel file e ti permetta di parlarci. Gli strumenti qui sotto coprono tutto: dalle app di chat a un click per principianti ai server industriali per team, più qualcuno che richiede una workstation con più memoria di quanto disco abbiano molti laptop. Scegli per categoria: un Runtime è il motore, una GUI è l'app amichevole sopra, un Server serve per condividerlo con più utenti. Dove una voce ha una soglia hardware, i suoi compromessi lo dicono per primi. Le schede sono in inglese (dati del catalogo). Se stai iniziando, Ollama è la porta d'ingresso più semplice.

Runtime3 voci

Ollama

Runtime

MIT

The fastest way to get a local LLM running with one command.

macOSLinuxWindows

Strengths

One-line install, one-line model pulls
Built-in OpenAI-compatible API on localhost:11434
Active model library with 4,500+ tagged variants

Trade-offs

Less raw throughput than vLLM under heavy concurrent load
Configuration is opinionated; advanced tuning means dropping into llama.cpp anyway

Visit project →Read install guide

llama.cpp

Runtime

MIT

Maximum control and the broadest hardware coverage in the open ecosystem.

macOSLinuxWindowsAndroidiOS

Strengths

Runs almost anywhere: CUDA, ROCm, Metal, Vulkan, CPU-only
Tight GGUF quantization control
Reference implementation behind most desktop LLM tools

Trade-offs

Command-line first; the UX assumes you read READMEs
Quantization options multiply quickly, easy to pick the wrong one

Visit project →Read install guide

DwarfStar (ds4)

Runtime

MIT

Running DeepSeek V4 Flash at usable speed on one 128 GB machine, and deliberately little else.

macOSLinux

Strengths

Asymmetric 2-bit quantization of the routed MoE experts only, which is what puts a 300B-class model inside 128 GB
Metal, CUDA and ROCm backends, with benchmarks published per machine rather than as a single headline number
On-disk KV cache, so a long session survives a restart

Trade-offs

Wants 96-128 GB of RAM to hold the Q2 build in memory, 256 GB or more for Q4; below that you are on the slower SSD-streaming path
Runs one model family on purpose: DeepSeek V4 Flash and PRO, plus GLM 5.2
The author calls it beta quality and the code changes fast

Visit project →No guide yet

Interfacce grafiche5 voci

LM Studio

GUI

Proprietary

Browsing, comparing and chatting with local models in a desktop GUI.

macOSLinuxWindows

Strengths

Polished chat UI with side-by-side model comparison
Built-in Hugging Face model browser
Local OpenAI-compatible API server with one click

Trade-offs

Closed source; the engine is llama.cpp but the shell is not
Less scriptable than CLI-first tools

Visit project →Read install guide

Open WebUI

GUI

MIT

A multi-user web frontend that talks to Ollama or any OpenAI-compatible backend.

Web (Docker), self-hosted

Strengths

Multi-user with authentication and chat history
Tool calling, RAG and prompt templates out of the box
Drop-in replacement for the ChatGPT web UI inside your network

Trade-offs

Needs Docker or Python plus a separate inference backend
Feature breadth means a steeper config surface

Visit project →No guide yet

Jan

GUI

AGPL-3.0

An open source desktop alternative to LM Studio.

macOSLinuxWindows

Strengths

Fully open source desktop client
Local-first design, no required cloud account
Plugin system for extensions

Trade-offs

Model catalog is smaller than LM Studio's
Newer project; some rough edges on Windows

Visit project →No guide yet

GPT4All

GUI

MIT

A friendly desktop client aimed at non-technical users.

macOSLinuxWindows

Strengths

Lowest barrier to entry of any desktop LLM client
Local document chat (RAG) built in
Cross-platform installers

Trade-offs

Less raw control than llama.cpp
Performance depends on the bundled engine version

Visit project →No guide yet

text-generation-webui

GUI

AGPL-3.0

Power users who want every knob exposed.

LinuxWindowsmacOS

Strengths

Supports multiple backends (Transformers, llama.cpp, ExLlamaV2)
Detailed sampler controls
Extension ecosystem for RAG, characters, voice

Trade-offs

Setup can be fiddly across CUDA versions
UI density is intimidating for newcomers

Visit project →No guide yet

Server3 voci

vLLM

Server

Apache

Production-grade inference with concurrent users and high throughput targets.

Linux (CUDA, ROCm)

Strengths

PagedAttention for memory-efficient KV cache
Continuous batching and speculative decoding
An order of magnitude more throughput than Ollama under heavy concurrency

Trade-offs

GPU-only path; not aimed at single-user desktops
Operational complexity is real; budget for tuning

Visit project →No guide yet

LocalAI

Server

MIT

A self-hosted drop-in for the OpenAI API, with multi-model support.

LinuxmacOSWindows (Docker)

Strengths

OpenAI API compatibility across chat, embeddings, images, audio
Pluggable backends including llama.cpp, whisper.cpp, diffusers
Designed for Docker and Kubernetes deployments

Trade-offs

Configuration sprawls quickly as you add modalities
Performance depends heavily on the underlying backend you pick

Visit project →No guide yet

Hugging Face Text Generation Inference

Server

Apache

A production server that pairs naturally with Hugging Face Hub.

Linux (CUDA, ROCm)

Strengths

Tensor parallelism, continuous batching, quantization
First-class integration with HF Hub models
OpenAI-compatible endpoint

Trade-offs

Throughput sometimes lags behind vLLM on the same hardware
Less community plugin work than vLLM

Visit project →No guide yet

Orchestratori1 voce

Kubernetes + Kubeflow

Orchestrator

Apache

Operating inference at scale across a fleet of GPUs.

Linux (any cluster)

Strengths

Mature operator pattern for batch and online inference
Pairs well with vLLM and Triton
Strong story for multi-tenant workloads

Trade-offs

Heavyweight; only worth it past a certain scale
Ops cost is non-trivial

Visit project →No guide yet

Framework2 voci

LangChain

Framework

MIT

Wiring models, tools, retrieval and memory into application logic.

PythonJavaScript

Strengths

Large community and integrations catalog
Useful patterns for agents, RAG, multi-step chains
Pairs with most local runtimes via OpenAI-compatible APIs

Trade-offs

Surface area is huge and not always cohesive
Abstraction overhead is sometimes more cost than value

Visit project →No guide yet

LlamaIndex

Framework

MIT

Building retrieval pipelines and document-grounded chatbots.

PythonTypeScript

Strengths

Strong primitives for indexing and retrieval
Many connectors to data sources
Works against any OpenAI-compatible local endpoint

Trade-offs

Naming and module reshuffling has been frequent
Some abstractions feel premature

Visit project →No guide yet