LocalAI hosting

A drop-in OpenAI replacement for text, images, audio and embeddings.

LLM runtimesMITfrom $16.99/mo

In short

LocalAI hosting at AIHostBox means a dedicated, single-tenant LocalAI instance on our infrastructure: we install it, pin it to a reviewed release, put it behind TLS on your domain, back it up nightly and keep it patched. Plans start at $16.99 per month and the admin account, the data and any API keys are yours.

OpenAI API surfaceEmbeddingsWhisperTTSMultiple backends

LocalAI is an MIT-licensed inference server created by Ettore Di Giacinto and maintained by a community team. Where Ollama focuses on chat models, LocalAI tries to be the whole OpenAI API in one binary: chat and completions, embeddings, audio transcription, text-to-speech and image generation, each served by a pluggable backend such as llama.cpp or whisper.cpp. Recent releases also answer Anthropic- and ElevenLabs-style requests, and ship a web UI for browsing and installing models from its gallery.

Running it on a server rather than a laptop is mostly about who else needs it. Once an endpoint has a stable URL and an API key, every tool in the building — a chat front-end, an automation, a RAG pipeline — can share the same models instead of each person keeping a copy. It also keeps loaded models warm, so the first request of the morning does not wait on a laptop waking from sleep.

LocalAI hosting plans

StarterRuns 7-8B models (Q4)

8 GB RAM · 4 vCPU · NVMe

$16.99/mo

Order

PlusRuns 13B models (Q4)

12 GB RAM · 6 vCPU · NVMe

$27.99/mo

Order

Every plan is a dedicated single-tenant instance with NVMe storage, nightly off-box backups, TLS and managed patching. Quarterly, semi-annual and annual terms are discounted at checkout. Model tokens are not included — you bring your own key.

How we run LocalAI

The CPU LocalAI container on port 8080 with persistent volumes for models, backends, configuration and generated files, as in the project’s own compose file; no separate database.

What we configure for you

LOCALAI_API_KEY set before first exposure, with one key per client app if you want
MODELS_PATH on persistent storage so models survive upgrades
LOCALAI_THREADS matched to the plan’s vCPU count
LOCALAI_CONTEXT_SIZE sized to what the plan’s memory can hold
LOCALAI_SINGLE_ACTIVE_BACKEND so only one large model sits in RAM at a time
Your first chat, embedding and Whisper models installed from the gallery

Which plan fits

Starter (8 GB, 4 vCPU) holds one 7-8B Q4 chat model plus a small embedding model, or Whisper for transcription. Plus (12 GB, 6 vCPU) runs 13B Q4 models or keeps a 7-8B model loaded alongside embeddings and Whisper.

The honest limits

Everything runs on CPU: chat sits at roughly 5-15 tokens/s on 7-8B Q4, image generation is slow, and models above 13B do not fit these plans.

What people run LocalAI for

Swap out a paid API

Change the base URL in an existing app and keep the same client code.

One endpoint, several modalities

Text, embeddings and transcription from a single service.

Private embeddings for RAG

Generate vectors for a document index without sending the documents to a third party.

Shared backend for chat front-ends

Point Open WebUI or LibreChat at LocalAI and give the team one set of models.

Licence and version policy

Item	Position
Upstream licence	MIT
Sold as managed hosting	Yes
Tenancy	Single tenant — one customer per instance
Patching	Critical advisories within 24 hours, high severity within 72

Our full position on licences, including the applications we decline to sell, is on the licensing page.

Reading on LocalAI

ComparisonOllama vs vLLM ComparisonOllama vs LM Studio ComparisonWhisper alternatives for self-hosted transcription Guide · Private LLMRunning an LLM on CPU: what actually works

Works well with LocalAI

More LLM runtimes we host

Questions

LocalAI hosting: your questions

Which models can I run?

Anything its backends support: GGUF language models through llama.cpp, Whisper models for transcription, embedding models and TTS voices. Most are one click from the built-in gallery, and models can also be pulled from Hugging Face or an OCI registry. On our CPU plans, 7-8B Q4 models are the practical chat size on Starter and 13B Q4 on Plus.

How is it different from Ollama?

Ollama is a tidy model runner for chat and embeddings with its own API plus an OpenAI-compatible layer. LocalAI covers more of the OpenAI surface — transcription, speech and images too — through separate backends. If you only need chat, Ollama is simpler; if one endpoint must answer several kinds of request, LocalAI fits better.

Is the API protected?

Yes. We set LOCALAI_API_KEY before the instance is reachable, so every request needs a bearer token, just as with OpenAI. You can hold several keys, for example one per application, and revoke one without disturbing the rest.

How fast is it on CPU?

For a 7-8B model at Q4, expect roughly 5-15 tokens per second on CPU. Embeddings and Whisper transcription are more forgiving and work well for batch jobs. Image generation runs on CPU but is slow enough that we would not build a product on it.

What happens to my models during an upgrade?

Models and their YAML configuration live in the models volume, separate from the container image. Upgrading LocalAI replaces the image and leaves downloaded models in place, and the volume is part of the nightly backup.

Run LocalAI without running a server

A dedicated LocalAI instance, deployed and maintained by us. Month to month, cancel whenever.

Browse the catalogue Talk to a human

LocalAI hosting

LocalAI hosting plans

How we run LocalAI

What we configure for you

Which plan fits

The honest limits

What people run LocalAI for

Swap out a paid API

One endpoint, several modalities

Private embeddings for RAG

Shared backend for chat front-ends

Licence and version policy

Reading on LocalAI

Works well with LocalAI

Open WebUI

LibreChat

AnythingLLM

Flowise

More LLM runtimes we host

Ollama

Hermes

KoboldCpp

llama.cpp

LocalAI hosting: your questions

Run LocalAI without running a server