am2Strata · a local AI server in Rust

What it does

One server on your machine, one address for every app. Your models, your data and your keys stay with you.

Compatible APIs any OS

OpenAI, Anthropic, Responses and Gemini-compatible endpoints, streaming and tools included, so existing apps and coding agents connect with just a base URL and a key.

Web app any OS

Chat with a live canvas for HTML, diagrams and Python in a sandbox; a Monitor for speed, GPUs and the queue; settings for everything else.

Workshop any OS

Convert a Hugging Face model to GGUF, quantize it, merge a LoRA adapter, and install the result. Fine-tune with LoRA, sized to the machine you have.

Training data any OS

Turn your documents into a chat dataset written by the model you have loaded, with near-duplicates dropped, an optional judge pass and a dataset card. Review and edit it record by record.

Sandbox VM Linux · KVM

Computer use in its own small Debian virtual machine: the agent installs and runs what it needs inside, and your PC is untouched. It pauses when idle and resumes in about a second.

Computer tab Linux

Watch the agent's desktop live, take control with your own mouse and keyboard, and copy and paste across. Rules you choose decide what it may do.

Secure by default any OS

HTTPS and secure WebSockets with a self-signed, your own or a Tailscale certificate. API keys with limits and expiry, and usage per key and per model.

One live connection any OS

The web app talks to the server over one WebSocket: answers stream over it, and status, the queue and jobs are pushed the moment they change.

Media build from source

Songs that sing their lyrics in 50+ languages, voices, images, video and speech to text, all running on your own GPU.

See it

Real screenshots of the web app, running on one PC.

A question and its answer, streamed at 68 tokens per second — **Chat.** Answers stream as they are written, with the model's thinking alongside.

A digital clock the model wrote, running in the canvas — **Canvas.** An HTML page the model wrote, running live beside the chat.

Python from the chat run in a sandbox with no network — **Python, sandboxed.** Code from the chat runs with no network and your home folder hidden.

A Mermaid flow and shapes drawn on the whiteboard — **Flows and whiteboard.** Diagrams open by themselves when an answer has one.

The Monitor's request history — **Monitor.** Every request, with speed, GPU, memory and what came back.

Settings, Connect: the coding agents and what each supports — **Coding agents.** Connect the agents you use to your own models in one command.

The Workshop and Computer tabs are newer than these screenshots.

Install in one line

The installer picks the right download for your system and processor, checks its SHA-256, needs no admin rights, and keeps your models and settings when you update.

curl -fsSL https://github.com/abhaymin/am2strata/releases/latest/download/install.sh | sh

1 · Get a model

Pull a small one to start:
am2strata pull mistralai/Ministral-3-3B-Instruct-2512-GGUF Q4_K_M

2 · Start it

Run am2strata and open http://127.0.0.1:8000 in your browser.

3 · Check your PC

am2strata-run doctor lists each optional feature and exactly what this machine still needs for it.

Where it runs

Six downloads. The server, the web app, the APIs and the Workshop work on all of them.

	Linux x64	Linux Arm	macOS Apple silicon	macOS Intel	Windows x64	Windows Arm
Server, web app, APIs, Workshop	yes	yes	yes	yes	yes	yes
Rust engine	CPU or GPU	CPU	CPU or GPU	CPU	CPU or GPU	CPU
Computer tab	yes	yes	–	–	–	–
Sandbox VM	yes (KVM)	–	–	–	–	–
Start at sign-in	systemd	systemd	launchd	launchd	Startup	Startup