LocalAI CLI Cheatsheet - local-ai Self-hosted Inference Server Full Reference

LocalAI is an open-source, self-hosted OpenAI-compatible inference server for LLM, embeddings, and audio that runs on CPU without a GPU. local-ai is its unified CLI: run and serve start the server, models manage weights, api exposes standard endpoints. This sheet covers launch, model management, and calls so you can self-host a private AI backend.

AI CLI·20 commands·Last updated 2026-09-10

Start Server 6

local-ai run

Start with default config

local-ai serve

Start the API server (same as run)

docker run -p 8080:8080 localai/localai

Run via Docker

local-ai start

Start resources from a file

local-ai --help

List all commands

local-ai serve --models-path ./models

Start with a custom model dir

Model Management 7

local-ai models list

List installed models

local-ai models install gpt4all

Install a gallery model

local-ai models pull gpt4all

Pull a model from the repo

local-ai models install llama-3-instruct

Install a named model

local-ai models apply

Apply model configuration

local-ai models list --help

models subcommand help

local-ai models install --help

Install flags

API and Finetune 7

local-ai api

Start the API service

curl http://localhost:8080/v1/chat/completions -d '{"model":"gpt4all","messages":[{"role":"user","content":"hi"}]}'

Call chat completions

curl http://localhost:8080/v1/models

List available models

local-ai finetune

Start a finetune job

local-ai api --help

api help

curl http://localhost:8080/v1/embeddings -d '{"input":"text"}'

Call embeddings

local-ai models apply --help

apply help

Tips

LocalAI is OpenAI-compatible; point an existing project base_url at port 8080 to go private.
It runs without a GPU: smaller quantized models still give basic completion on CPU.
models install pulls from the gallery; put custom weights in --models-path for easier management.

FAQ

How to choose between LocalAI and Ollama?

LocalAI focuses on OpenAI-compatible self-hosting (with embeddings and audio) and richer config; Ollama is lighter and quicker to start. Both run local models.

Can it run without a GPU?

Yes. LocalAI supports pure CPU inference, slower but workable with small quantized models.

Official References

Each command links to its official documentation below, so you can verify the latest usage and read deeper.

LocalAI Docs

Maintained by LaoHand

Publicly updated on Sep 10, 2026, continuously proofread against official docs.

Contact Us

Wrong command or description? Send us corrections, business inquiries or product feedback by email.