Frequently asked questions

Aistatine @code is an on-device AI coding agent that installs locally on your machine. It helps you write, debug, and refactor code while keeping the workflow on your device rather than sending repository content to a hosted AI service.

@code is designed for local inference: its public installer downloads the model to your machine and the tool runs through a local llama.cpp server. Evaluate its fit against your own privacy and deployment requirements.

No. Because everything runs locally, your code never reaches our servers. We have no access to your prompts, your codebase, or your outputs — so there is nothing to train on.

A 1-million-token context window means @code can read and reason over your entire codebase at once — not just one file. It understands cross-file dependencies, architecture patterns, and subtle bugs that smaller-context tools miss.

Slash commands are a built-in shortcut system in @code. Type / followed by a command name to trigger specific modes — for example /web to control web interactions, /build for end-to-end app building, /knowledge for data retrieval, and /effort to switch between super and fast modes.

No. Once the model is downloaded on first run, Aistatine works fully offline. You can code on a plane, in a secure facility, or anywhere without internet access.

There are no per-token API costs. Once installed, @code runs inference locally using your own hardware. You pay nothing for each prompt or code generation — run it as much as you need.

Aistatine @code supports macOS, Linux, and Windows. Install with a single terminal command on any platform and be up and running in minutes.

On macOS or Linux, open your terminal and run:

curl -fsSL https://raw.githubusercontent.com/aslam-anees/at-code/src/install/install.sh | bash

On Windows PowerShell, use the PowerShell command (do not use curl, which is an alias for a different command there):

Invoke-RestMethod https://raw.githubusercontent.com/aslam-anees/at-code/src/install/install.ps1 | Invoke-Expression

The installer sets up the @ command mapping and downloads the model on first launch.

The public installation guide documents 8 GB RAM as the minimum and about 4.8 GB free disk space. Supported systems are macOS 12+, Windows 10+ 64-bit, and a modern Linux distribution. The installer downloads the local @0.1-9b.gguf model and its multimodal projection file (about 4.8 GB total) on first run.

The public installation guide states that @code can work offline after the initial binary and model download. For an air-gapped environment, obtain, validate, and stage the installer, binary, and model files through your organization’s approved software process before disconnecting the machine. Aistatine does not currently publish a separate air-gap deployment guide.

@code is designed to run the model and coding workflow locally on the device where it is installed. That means repository files and prompts remain in that environment during local inference. Your organization remains responsible for device storage, backups, source control, endpoint security, and any tools or commands it chooses to run alongside @code.

Aistatine does not currently publish enterprise license terms, centralized audit-log behavior, or an enterprise administration guide. Do not assume those capabilities are available. Contact the team with your deployment and compliance requirements so they can be evaluated against the current product.

The public documentation does not yet publish a supported-language or framework matrix. Evaluate @code with a representative repository, your normal test suite, formatter, and static-analysis tools before adopting it for a production stack. Aistatine will publish a compatibility matrix when it is documented and tested.