Local AI Lab

Terminal-first control center for a private local AI workstation.

Models, coding agents, web search, workspace knowledge, images, and monitoring. No paid API required.

By Bogdan Shchavinskyi at bogdanlabs.dev.

Install

Download and run the native installer for your host. It is a self-contained Go binary with no runtime dependency.

Linux amd64:

curl -fsSL https://github.com/master-bogdan/local-ai-lab/releases/latest/download/local-ai-lab-installer_linux_amd64 -o /tmp/local-ai-lab-installer
chmod 700 /tmp/local-ai-lab-installer
/tmp/local-ai-lab-installer

Linux ARM64:

curl -fsSL https://github.com/master-bogdan/local-ai-lab/releases/latest/download/local-ai-lab-installer_linux_arm64 -o /tmp/local-ai-lab-installer
chmod 700 /tmp/local-ai-lab-installer
/tmp/local-ai-lab-installer

Apple Silicon macOS:

curl -fsSL https://github.com/master-bogdan/local-ai-lab/releases/latest/download/local-ai-lab-installer_darwin_arm64 -o /tmp/local-ai-lab-installer
chmod 700 /tmp/local-ai-lab-installer
/tmp/local-ai-lab-installer

The installer shows the exact version, platform, application path, and command path before writing. It verifies SHA-256 checksums and also verifies GitHub release integrity and build provenance when gh is available. It installs only the control center; no model or service starts automatically.

Launch it from any directory:

local-ai-lab

Requirements

Host	Minimum	Status
Linux amd64 with NVIDIA or AMD GPU	6 GiB VRAM, 16 GiB RAM	supported
Linux amd64 with Vulkan-capable AMD or Intel GPU	6 GiB VRAM, 16 GiB RAM	experimental
Linux ARM64 with a supported GPU runtime	service-dependent	experimental
Apple Silicon macOS	24 GiB unified memory	supported

Fedora, Ubuntu, and Arch Linux are supported. CPU-only systems, Intel Macs, Windows, and WSL are rejected. GPU drivers must already work on the host. Linux setup can install Docker and supported container runtime packages after showing exact commands and receiving confirmation. macOS requires Homebrew and Docker Desktop.

Use

First run checks hardware and disk, recommends models for the detected memory, and lets you confirm every model, service, path, and system change. Installation ends with all services stopped. Start the workload you need from the control center afterward.

Main actions:

• start or switch coding, image, infrastructure, or combined workloads
• inspect service health and localhost URLs
• follow logs and manage Ollama models
• configure ComfyUI, monitoring, and OpenCode
• index a Git workspace for local agent search
• update or roll back the Local AI Lab application
• stop services or uninstall with a reviewed deletion plan

Services can remain running after the TUI exits. Exit flow asks whether to keep them running or stop them. Reopen local-ai-lab later to inspect or stop them.

Included Tools

Tool	Purpose
Ollama	local model runtime
OpenCode	terminal coding agent; installed separately, configured by the TUI
Open WebUI	browser chat and local RAG
SearXNG	local web-search gateway
Qdrant	workspace vector search
ComfyUI	separate image-generation flow
Prometheus, Grafana, cAdvisor	predefined metrics, alerts, and dashboard

Model picker labels each model FAST, TIGHT, HYBRID, or UNSUPPORTED, explains why, and sizes context for available memory. Standard models are shown first. A separate uncensored catalog contains a small set of community abliterated models; these are never selected by a preset, have unverified tool calling, and require explicit install confirmation. Their exact Hugging Face revisions and SHA-256 checksums are pinned.

A preserved reinstall receipt can restore model and service choices after a full uninstall, but never stores secrets, prompts, search history, logs, or indexed workspace paths.

Local Data

Default root:

• Linux: ${XDG_DATA_HOME:-~/.local/share}/local-ai-lab
• macOS: ~/Library/Application Support/local-ai-lab

Application versions, configuration, models, indexes, cached searches, generated images, service data, and monitoring history remain local. Services bind to 127.0.0.1; other processes running as your user can still access localhost APIs. Web searches and downloads use the internet.

Choose Application -> Uninstall for application-only removal, full removal with reinstall choices preserved, or absolute removal. Docker, GPU drivers, Homebrew, native apps, and shared upstream container images are never removed.

Development

From a source checkout:

make start
make test

make start builds a repository-local development binary and uses the same lab data as an installed release. It does not install a global command.

License

Local AI Lab source and configuration are MIT licensed. See LICENSE. Downloaded models, containers, applications, and dependencies retain their upstream licenses.