Skip to content

What is AI:GO?

AI:GO is a cross-platform desktop application and personal AI runtime for macOS, Windows, and Linux. It packages local inference servers, model discovery, workspace agents, document grounding, and an OpenAI-compatible API behind one interface so you can run models on your own hardware without compiling engines or managing server processes by hand.

The Problem with Cloud AI

Relying exclusively on hosted AI services such as ChatGPT or Claude introduces practical constraints for everyday technical and professional work.

  • Privacy: Prompts, source code, and attached documents travel to external servers outside your network boundary.
  • Recurring cost: Heavy interactive use or automated workflows rack up monthly seat fees and per-token charges.
  • Network dependency: Flights, air-gapped labs, and outages cut off access to your assistant.
  • Model drift: Hosted endpoints update, deprecate, or swap model checkpoints and sampling behavior on the provider's schedule rather than yours.

The AI:GO Solution

AI:GO runs open-weight language, vision, speech, and image models directly on your computer, while still letting you attach cloud APIs or other machines when a task calls for a larger model.

High-Performance Inference

AI:GO spawns and supervises dedicated inference backends matched to your hardware, so a model crash never brings down the app. On laptops and workstations it runs llama.cpp (Metal, CUDA, ROCm, Vulkan, CPU), Apple MLX, Apple's Built-in Mac Model (Apple Foundation Model on macOS 26 or later with Apple Intelligence), stable-diffusion.cpp and MLX image pipelines for image generation, and whisper.cpp for speech-to-text, plus container engines such as vLLM and SGLang for high-throughput serving.

Unified Experience

One Models window handles Hugging Face search and downloads with live hardware-fit checks, while Additional model folders reads existing LM Studio, Ollama, and Hugging Face cache directories in place without duplicating gigabytes on disk. Local models, cloud providers, and peer-to-peer Network nodes all route through a single OpenAI-compatible layer, letting you switch models mid-conversation without changing tools.

Beyond Chat: Agents & MCP

Alongside standard chat, AI:GO includes agent workflows that act on local files, schedules, and connected services under explicit permission controls. You can attach a project folder in Cowork for multi-step coding and file tasks, coordinate multi-agent teams in Squad, index reference documents into Data Hub & Wiki, schedule recurring Automations, surface upcoming events and tasks with Glance, connect Calendar, Reminders, Email, and Contacts, and plug in external tools over the Model Context Protocol (MCP).

AI:GO vs. Backend.AI

AI:GO and Backend.AI address different scales of AI infrastructure and work together when needed.

What is Backend.AI?

Backend.AI (Core and Enterprise) is a cluster-scale AI Infrastructure Operating System built for enterprises, cloud providers, and research centers.

  • Cluster scheduling: Pools and partitions hundreds or thousands of GPUs across multi-node clusters.
  • Multi-tenant isolation: Serves teams concurrently with quotas, container isolation, and role-based governance.
  • Training and production serving: Runs distributed training, fine-tuning, batch pipelines, and high-availability model endpoints.

AI:GO's Role

AI:GO is the personal AI desktop client and local server that runs on a single laptop, workstation, or home server.

  • On-device execution: Runs quantized and full-precision models directly on your Mac or PC with zero cluster setup.
  • Ready-to-use workspace: Pairs chat, image generation, translation, Cowork, Squad, and local Data Hub search in one desktop app.
  • Cluster bridge: Connects to a Backend.AI deployment or peer AI:GO nodes when a job needs frontier-scale models such as Solar-Open-100B or Qwen3.5-122B-A10B, keeping the same chat and agent interface.

Who is it for?

  • Developers: Run a local OpenAI-compatible endpoint for coding assistants, test tool-calling schemas, and iterate with Cowork against local repositories.
  • Researchers: Compare quantizations, context lengths, and model families side by side with reproducible local benchmarks.
  • Writers and analysts: Draft, translate documents with glossaries, and query private PDFs in Data Hub without per-token billing.
  • Security-sensitive teams: Keep confidential documents and source code inside a local machine or air-gapped network.

Continue to the Quickstart Guide to install AI:GO and run your first model.