Launch offer — 40% off every tier for a limited time. Prices shown are already reduced.

Local AI

Atlas
Self-hosted AI

Local model runners, vector stores and RAG frameworks wired into a private inference stack.

OpenAI-compatible API on your hardware
Vector search and RAG ready to wire
Prompts and data stay local

In the box

A working, pinned Docker Compose stack for self-hosted AI. Not a link dump - the integration layer: services wired together, environment template, a quickstart that reaches a first completion, a runbook for operating it, and a script that fetches the exact upstream sources at pinned commits.

16 components 23 files 5 tiers Docker Compose

Best for

Builders who need inference without sending data out.

Pricing

Choose your tier

One-time purchase. Higher tiers add components; the top tier adds a live onboarding session. Every tier includes the runbook, verify script and component guides.

Starter

The entry point. Two or three services, one compose file.

$12 ~~$20~~one-time 40% off

A working Compose stack for local inference
Quickstart that gets you to a first completion
Pinned source-fetch script for all three projects

Get Atlas Starter

3 components · instant download

Growth

Adds the core services most teams reach for next.

$30 ~~$50~~one-time 40% off

Everything in Starter
Three vector engines, running and wired
Retrieval smoke test proving the index works

Get Atlas Growth

6 components · instant download

Pro

The working set - most of the stack, documented end to end.

$60 ~~$100~~one-time 40% off

Everything in Vector Search
Four RAG and agent frameworks, documented
End-to-end architecture guide

Get Atlas Pro

10 components · instant download

How it fits together

Each service runs as its own container. The compose file defines the network, the volumes and the wiring between them.

Components

16 services, pinned

Every project keeps its own licence and copyright. The stack ships the integration, not a fork.

Component	What it does	Licence	Source
ollama Go	Local model runner exposing an OpenAI-compatible API.	MIT	upstream 1abe35e6e6
llama.cpp C++	Direct GGUF inference engine	MIT	upstream 0c1e57098b
text-generation-inference Python	Production Hugging Face model serving	Apache 2.0	upstream b4adbf2f6e
vllm Python	High-throughput LLM serving for GPUs.	Apache 2.0	upstream 36229e1337
LocalAI Go	Drop-in OpenAI replacement for local models.	MIT	upstream 62e0827cba
qdrant Rust	Vector database for similarity search.	Apache 2.0	upstream 6ab21cac18
chroma Rust	Embedded vector store for retrieval.	Apache 2.0	upstream 848cf58dd6
milvus Go	Distributed vector database.	Apache 2.0	upstream 1211f93389
faiss C++	Similarity search library for dense vectors.	MIT	upstream 88a28bcd80
llama_index Python	RAG and document pipeline framework	MIT	upstream 7e2c60a78e
langchain Python	Framework for building LLM applications.	MIT	upstream 026c3da2b6
haystack Python	Production NLP and RAG pipelines.	Apache 2.0	upstream 0499239e66
crewAI Python	Multi-agent orchestration framework.	MIT	upstream ca1d55abb3
aider Python	Terminal AI pair programmer.	Apache 2.0	upstream 5dc9490bb3
semantic-kernel C#	Microsoft's LLM orchestration SDK	MIT	upstream 58ad1a3780
continue TypeScript	Open-source IDE coding agent.	Apache 2.0	upstream 5522c6f44c

Documentation

A guide for every component

Written docs that ship inside the archive - role, reasoning and how to run each service.

aider chroma continue crewai faiss haystack langchain llamacpp llamaindex localai milvus ollama qdrant semantickernel tgi vllm

Quickstart

Run it

# 1. fetch the pinned upstream sources
./fetch-sources.sh ./sources

# 2. configure and start
cp .env.example .env
docker compose up -d

# 3. confirm everything came up
./verify.sh

Docker Engine 24+ with the Compose v2 plugin. 4 GB RAM for light tiers, 16 GB+ for the complete deployment.

Atlas Self-hosted AI