Switch models freely
Change models mid-conversation and compare answers. Any OpenAI-compatible service or local model works.
AI workspace
Atlas Chat
Many AI models in one interface, shared by your team, on your server.
Atlas Chat is a self-hosted AI chat workspace. OpenAI, Anthropic, Google, DeepSeek and models running on your own hardware are all available from one interface. Conversations, uploads and prompt presets stay on your server, and each team member signs in to their own history.
Try it now
This interactive demo runs in your browser with fictional sample data, and your changes stay in your browser. Download the kit for the complete system.
Core features
For teams that want one AI entry point for the whole company while keeping control of data and spend.
Files and images
Upload PDFs, spreadsheets and images to review contracts, read reports or describe pictures.
Agents and presets
Bundle prompts, reference files and tools into shared agents such as “Polish a sales email” or “Check a quotation”.
Conversation management
Full-text search, branching, edit-and-regenerate, export and sharing, isolated per user.
Accounts and access
Email sign-in with optional closed registration, and single sign-on for company identity.
Keys and usage
Use company-wide keys or let users bring their own, with usage limits to keep costs predictable.
How teams use it
- 01
Sales assistant
Reps translate inquiries, polish emails and summarize requirements from one entry point, without separate subscriptions.
- 02
Internal Q&A
Upload product manuals and spec sheets so new staff can ask for parameters directly.
- 03
Local models
Sensitive material is handled by a model on the company server, so data stays inside the network.
Download and deploy
The kit contains the service definition, a settings template, a one-command installer and a guide. The installer generates secrets, prepares the database and waits until the service is ready.
Download the kit
atlas-chat.zip
Start in three steps
- Unzip the downloaded kit
unzip atlas-chat.zip - Open a terminal in the folder and run the installer
cd atlas-chat && sh install.sh - Open in your browser
http://localhost:3030
First run
Register an account (the first one becomes admin). Add model keys to the settings file, or leave user_provided so each user enters their own; local models are configured in atlas.yaml.
Requirements
- Docker
- Docker 24+ with Docker Compose v2
- Free memory
- ≥ 3 GB
- Disk space
- ≥ 5 GB
- OS
- macOS, Linux, or WSL2 on Windows
Services
webChat UI and APImongodbUsers and conversationsmeilisearchFull-text conversation search
What's in the kit
- Web chat interface with Chinese UI
- Multi-provider setup and local-model template
- File upload and full-text search
- Agents and prompt presets
- One-command installer (with database and search)
What we can tailor for you
For production rollout, connections to your existing systems or brand customization, we can take it from here.
Let’s talk- Your existing model accounts or private models
- Agents and prompt libraries for your workflows
- Single sign-on and per-user limits
- Your name, logo and welcome message
FAQ
- Do I need overseas model accounts?
- No. You can use only regional model services or open models on your own server.
- Will providers train on our chats?
- Chat history is stored only on your server. Content sent to a provider follows that provider's API terms; use a local model for sensitive work.
- How many people can use it?
- That depends on your server and the model provider's rate limits; the kit has no user cap.
- Can I close public registration?
- Yes. Set ALLOW_REGISTRATION to false and have an admin add accounts.