Your own AI, on your own hardware.

Run open models for chat, images, speech and video on your own machines. The API speaks the OpenAI and Anthropic formats. Your team gets chats, files and documents on top, and the assistant can write and run code.

curl -sSL https://localai.io/install.sh | sh

Test build from the development release. The release build is coming.

Windows, Android and iPhone: coming soon.

Or a container, the desktop app or Kubernetes. Then open http://localhost:8080.

42 seconds. We made this film with LocalAI: the pictures, the voices and the music.

CPU or GPUNVIDIA, AMD, Intel and Apple
18 engines we wroteC++ and GGML, with no Python at inference
OpenAI-compatibleAlso Anthropic and ElevenLabs formats
MIT licenceOpen since March 2023, 50,388 stars

One API

Change the base URL. Keep your client.

Any OpenAI client works. The same server also does speech, images, video and music.

python
1 from openai import OpenAI
2 client = OpenAI(
      base_url="http://localhost:8080/v1",
      api_key="none")
3 chat = client.chat.completions.create(
      model="gemma-4-e2b-it-qat-q4_0",
      messages=[{"role": "user", "content": "Hello"}])
4 print(chat.choices[0].message.content)
5 text = client.audio.transcriptions.create(
      model="whisper-1", file=open("meeting.mp3", "rb")).text
6 print(text)

Both models come from the gallery. For the second call, install whisper-1 from the gallery.

Every kind of model

Text, images, video, music and voice.

Install a model from the gallery and call it from your code.

  • Text and vision: chat, tool calls, image understanding, object detection.
  • Generate images, video and music. With the Sandbox (development build), the assistant can also make short MP4 videos with ffmpeg, such as a title card or a slideshow. Look at the result before you use it.
  • Speech to text, text to speech, embeddings and a reranker.

Workspace

Give it your files. Get documents back.

Chat with a model, give it your files and get documents back.

  • Your files stay on your server. Share folders, and undo any change.
  • Ask in one sentence, get a Word, Excel, PowerPoint or PDF file.
  • A Sandbox runs Python on your files. It has no network by default.
  • Knowledge searches your documents by meaning.

Office and Sandbox are add-ons. An admin installs them once.

Team

One server. Your whole team.

Each person has a login, chats and files. An admin chooses what each person can use.

  • Share folders, saved agents and Knowledge collections with one person or with everyone.
  • Turn features on or off for each person.

Projects

Keep the work, its files and its notes together.

A project groups chats around one folder, with files and notes that its chats read.

  • The model proposes a note. You confirm it before it is saved.
  • Make an agent from a project, or share the folder with a teammate.
The Projects page of LocalAI, showing a project with its chats, files and notes.

Models and hardware

Open models, your hardware.

LocalAI installs the model and downloads the backend for your CPU or GPU.

  • Runs on a plain CPU. NVIDIA, AMD, Intel and Apple GPUs make it faster.
  • Run it on a laptop, a server, in a container or on Kubernetes. Distributed mode spreads models over several machines.
  • More than 80 backends, from llama.cpp to vLLM. We wrote 18 of them.

Get started

Install with one command.

Install a server with a script, Docker or Kubernetes, or use the desktop app. Backends download the first time a model needs one.

Desktop app

Test build from the development release. The release build is coming.

Windows, Android and iPhone: coming soon.

macOS: a DMG for Apple silicon. Linux: an AppImage for x86-64. For ARM64 Linux, take LocalAI-Launcher-dev-arm64.AppImage from the development release. Each file has a .sha256 file. Desktop app guide

The LocalAI desktop app on first start. It shows the model that it recommends for this machine, with an Install button.
shell

Script

$ curl -sSL https://localai.io/install.sh | sh
$ local-ai
# open http://localhost:8080 and install a model from there

More: containers and GPU images, the desktop app (Linux and macOS), Linux, Helm chart, build from source

The install script and the latest image give 4.12.1 today. Projects, Office, Sandbox, Team, add-ons and the new desktop app need a development build or a newer release.

One command to install.

localai.io