CryoZen
Self-hosted AI workspace

Your AI workspace. On hardware you control.

CryoZen runs chat, autonomous agents, document search, and a mail, calendar, and contacts client as one application on your own machine. Point it at local models or your own API keys. Nothing leaves unless you configure an integration that sends it there.

Available for macOS, Windows, Linux, and Docker.

127.0.0.1:7860

Summarize the three contracts in /legal and flag renewal dates.

Indexed 3 documents locally. Two auto-renew on 01 Mar; one needs 60 days notice.

Draft the notice email and hold it for my review.

local model · llama-3.1-70b · running on your GPU
Private by construction

Not a promise in a policy. A property of where it runs.

CryoZen is a server you start on your own hardware. The guarantees below are structural, not contractual.

Binds to localhost by default

The server listens on 127.0.0.1. Nothing is reachable from your network until you decide it should be.

Credentials encrypted at rest

API keys, mail passwords, and OAuth tokens are encrypted in your own database, on your own disk.

No telemetry

No analytics, no phone-home, no usage beacons. There is no build of CryoZen that reports on you.

Data leaves only when you say so

Connect a hosted model or an external mailbox and traffic goes there. Connect nothing and nothing leaves.

What it does

Everything below ships in the application today.

Not a roadmap. These are the surfaces you get on first launch.

Chat with any model

Connect local backends (Ollama, llama.cpp, vLLM, LM Studio) or hosted providers (OpenAI, Anthropic, Groq, DeepSeek, GitHub Copilot). Compare answers from several models side by side.

Agents with real tools

Give a model shell access, file access, web search, and MCP servers so it can finish multi-step work, inside a workspace boundary that confines what it can touch.

Research that cites its sources

An engine that searches, crawls, and synthesizes a cited report, backed by its own SearXNG instance so queries never reach a third-party search engine.

Your documents, indexed locally

PDFs, Office files, and text go into a local vector store for retrieval-augmented generation, plus long-term semantic memory that carries across conversations.

Mail, calendar, and contacts

A full IMAP and SMTP client with scheduled sending and reminders, CalDAV and CardDAV sync, tasks, and automations that run on a schedule or fire a webhook.

Serve models on your own hardware

Scan GPU and memory, get quantization recommendations that actually fit, and start or stop model servers, including over SSH on a remote host.

Who it is for

People who want the capability without handing over the data.

The same install scales from one laptop to a shared instance behind your own proxy.

Individuals

One private workspace for chat, research, mail, and notes.

Teams

A shared instance behind your own reverse proxy, with per-user accounts and per-user data.

Developers

An agent that runs shell commands, reads and writes files, searches the web, and calls MCP servers under your supervision.

Run it tonight on the machine in front of you.

Download a build, or clone the repository and bring it up with Docker Compose in three commands.

Every build is published on GitHub Releases.