Approve the actual action
Decisions shows runs, exact tool inputs and results. Code add-ons and MCP calls always need approval; calendar actions follow your autonomy level and the action’s risk.
Self-hosted · Open-Core · Free
Your own AI engine room.
Your self-hosted AI workspace and company OS: chat, CRM, calendar and agent tools in one place. With persistent runs, exact-input approvals and MCP connections. Choose local models or cloud AI — no dedicated GPU required. You decide what your AI is allowed to execute.
AI that acts. You stay in control.
From personal assistant to company workspace: cockpit, CRM, marketing and metrics meet a traceable tool runtime.
Runs with a memory
Run status, approvals and audit events persist in SQLite. Actions already started are not blindly replayed after a restart. After approving a tool, you start the next chat message yourself.
Connect MCP deliberately
Connect one MCP server over Streamable HTTP. Bearer authentication, an explicit tool allowlist and approval of exact inputs restrict what your AI can execute.
Your company workspace
Cockpit, CRM, marketing, metrics and optimization complement chat, calendar and documents. Company context stays pinned throughout each request.
Process boundaries for add-ons
Code add-ons run in short-lived Node processes with time, memory and output limits. These contain failures but do not replace container isolation for untrusted code.
API & TypeScript SDK
Query runs through authenticated HTTP endpoints and integrate with a lightweight TypeScript client. Run locally with Docker or on your server with Dokploy.
Everything you need — under your own roof.
Your everyday tools share one engine room — now joined by cockpit, CRM, marketing and metrics for your company. Explore a selection of the workspace apps.
Assistant
Streaming chat that shows you when it’s thinking. Pick the personality that fits you.
Vault
End-to-end encryption (AES-256-GCM) for messages and files. Shareable via messenger or email.
Connections
Email, calendar, Trello, MCP, cloud AI and local models — deliberately connected.
Inbox
Email over IMAP in your workspace. Overview, preview and replies in one place.
Documents
Store, manage and download your documents securely.
Research
Web research, bundled and sourced by the AI.
Studio
Image analysis with a local vision model. No image leaves the server.
Calendar
Your schedule at a glance, manageable by the AI.
Team
Personalities and your own company with roles and sub-agents that get work done.
How it works.
01
Install
Astoris runs anywhere — from an old laptop with cloud AI, to Ollama on a Mac, to your own GPU. One command and your engine room is up.
02
Connect
Hook up email, calendar, Trello, cloud AI or local models. For each connection you grant exactly what the AI may do.
03
Get started
Chat, encrypt, research, delegate to your agent team. All in one workspace, all on your side.
Security & sovereignty
Sovereign means: yours stays yours.
Astoris runs behind your own network. You decide whether local models or cloud AI answer — and either way you stay in control.
- HTTPS for access, end-to-end encryption in the Vault
- Sign in by password or Tailscale
- Encrypted SQLite contents and persistent approvals
- Local models (vLLM/Ollama) or cloud (Anthropic/OpenAI)
Behind your network
No open port to the internet required. Reachable over your private Tailscale mesh.
Tailscale login
Authentication over your mesh — only your devices get in.
Local or cloud
You choose per task: a private local model or cloud AI. Switchable any time.
One data point, not a requirement
Fast enough that it feels like mind-reading.
0 Tokens / second
An example from the high end: the chat model qwen3.5-35b in a power-user setup, measured on an NVIDIA DGX Spark (GB10). You don’t need this hardware — a cloud provider with a free tier or Ollama on a Mac is plenty for fluid streaming.
Vault
Encrypt. Share. You keep the key.
Messages and files are encrypted with AES-256-GCM before they ever leave your server. Share them over messenger or email — only whoever holds the key can decrypt them. No one in between, not even us.
- Client-side encryption
- For messages and files
- Shareable via messenger or email
What you should know about Astoris.
Short and factual — the key questions on cost, hardware, models and privacy.
What is Astoris?
Astoris is a self-hosted AI workspace and company OS. Assistant, Vault, Calendar and Documents are joined by cockpit, CRM, marketing, metrics and Decisions. The tool runtime persists runs and approvals. Choose local models or cloud AI.
How much does Astoris cost?
Astoris is free and open source (Open-Core, licensed under FSL-1.1-MIT). You can install and run it with no license fees. Some premium add-ons plug into paid third-party services (usually with a free tier) — you pay those costs directly to the provider, not to Astoris.
Do I need an expensive GPU?
No. Astoris also runs without your own GPU. You can use cloud AI providers with a free tier (e.g. Groq, OpenRouter, DeepSeek) or run local models via Ollama on a normal computer or Mac. A GPU speeds up local models but is optional.
Which AI models can I use?
It is your choice: local models via Ollama or vLLM running on your own hardware, or cloud providers such as Groq, OpenRouter, DeepSeek, Anthropic (Claude) and OpenAI. For each task you decide which model answers — switchable any time.
Is my data secure and truly self-hosted?
Astoris runs on your server, optionally within a private Tailscale network. The Vault uses end-to-end encryption; SQLite encrypts sensitive document contents and session data. Metadata and retained legacy JSON files are not fully covered. Cloud AI, MCP and external connections receive the data sent to them, even when your chat model runs locally.
Which apps and add-ons are there?
Classic workspace apps are joined by company tools such as cockpit, CRM, marketing and metrics, plus Decisions for run status and approvals. The catalog offers over 25 add-ons. Code add-ons execute in supervised processes and require your approval for every call.
How do I install Astoris?
Astoris is Open-Core and available on GitHub. You grab the project and start it on your own hardware — from an old laptop with cloud AI, to Ollama on a Mac, to your own GPU. The installation guide is in the GitHub repository.
Ready?
Build your own engine room.
Self-hosted, encrypted, sovereign — and free. Open-Core: grab it and keep control of your AI.
Questions? info@astoris.org