Private AI
Self-Hosted

Your private AI. Your hardware.Your rules.

Run private Ollama chat on your own GPU, NAS, or home server. Core local use needs no subscription or telemetry; voice, images, and remote services are optional operator integrations.

Free forever for local use. No credit card required.

atlas-ai terminal
A
Show me my AI models available locally.
Running on Ollama + llama3.2 • Chat: Llama 3.2 (3B) • Vision: Llama 3.2 Vision • Embeddings: all-MiniLM-L12-v2
GPU: RTX 4070 · 8.2 GBVRAM · 58°C
What makes it different

Five things the cloudcan't offer.

Local-first chat with durable memory and documents. Optional services stay explicit and operator-controlled.

Local-First

Core chat, prompts, and history stay on hardware you own. Optional network integrations run only when your operator enables them.

Persistent Memory

Save project facts, client context, and preferences in your local database. Relevant memories can be recalled in ordinary and agent chat.

Document Knowledge Base
Preview

Organize PDFs, DOCX, Markdown, or text in local knowledge bases. Chunking and Ollama embeddings are available when RAG is configured.

Optional Voice Mode

Install the accepted local whisper.cpp and Piper stack for private transcription, speech playback, and push-to-talk voice chat.

Self-Hosted Setup

Deploy with Docker Compose today. Windows installer source and a Linux helper are included, while signed release artifacts remain forthcoming.

How It Works

From download to AI in minutes

Docker handles the core services; onboarding guides model selection and reports optional integrations honestly.

01
Deploy Atlas

Start with the supported Docker Compose deployment on Windows or Linux. It brings up Atlas, PostgreSQL, migrations, and the private reverse proxy.

02
Connect Ollama

Point Atlas at an Ollama service and choose an installed model. Onboarding distinguishes stopped, unreachable, empty, and incompatible setups.

03
Save memory + upload docs

Save durable facts and organize documents locally. Enable RAG plus an Ollama embedding model when you want chunking and semantic retrieval.

04
Chat privately

Use the local dashboard for streamed chat and saved history. Optional agents, voice, images, plugins, and remote features expose their own setup gates.

Why local

The cloud is convenient.
Your data is the price.

Every hosted AI assistant trades your prompts, files, and habits for a smoother signup. Atlas flips that: core chat runs against your Ollama service and stores history in your PostgreSQL database. Optional network integrations are disabled until you enable them.

TopicHosted AIAtlas
Where the data livesOn someone else’s serversOn a machine you own
Conversation historyRetained by the vendorStays in your local database
Price modelMonthly subscription — usage caps, tier upsellsFree, forever. Bring your own hardware.
Offline / LAN useRequires a stable internet connectionCore chat works on your LAN when Ollama is local
Voice + image modelsSlow, gated, and count against your rate limitConnect local Whisper, Piper, and image backends when configured
Memory + uploaded documentsIndexed and stored on a third-party serverStored locally; embedding runs only when RAG is enabled
Tied to a vendorLock-in, model switches that break habitsChoose among models installed in your Ollama service

Atlas is open where it matters: bring your own models, swap backends, export everything. The product is yours in the same way the hardware is.

Early access

Be there at the launch.

Join a few thousand builders who want their AI off the cloud. Drop your email — we’ll send a single message when Atlas is ready to install.

One email at launch. No marketing drip, no upsells.

Looking for the installer already? Try the local chat dashboard.

Ready to take control of your AI?

Bring your AI home with the free local workspace. No credit card is required for core chat.

Want the details? Read the privacy policy.