QueAI | Modular AI Orchestrator. Your Stack.

// Modules

Composable modules.
Local or remote — your call.

Each module is a Docker container with its own UI, REST API and configuration. Some run models locally on CPU. Others can be thin proxies to public APIs (OpenAI, Anthropic, ElevenLabs). The kernel doesn't care — it routes and audits both the same.

Available

OCR

Optical Character Recognition

Tesseract pipeline on CPU. PDF, PNG, JPG and TIFF in, plain text out. Async job queue backed by Redis, server-sent events for live progress.

TesseractRedis + RQSSE

Available

STT

Speech-to-Text

faster-whisper on CPU with VAD filtering and configurable compute type. Audio in, transcript with optional timestamps out. Task queue, downloadable results, no audio leaves the host.

faster-whisperint8 CPUVAD

Available

TTS

Text-to-Speech

Piper voices on CPU with sub-second latency. English and Spanish multi-speaker out of the box. Streaming-friendly WAV output for agents, screen readers and bots.

PiperEN / ESlow-latency

Available

PLUGIN SDK

Build any module

A manifest.json and a docker-compose. Anything that runs in a container is a valid plugin — a CPU model, a GPU model, a thin proxy to a cloud API, or a pipeline that chains several together.

FastAPIDocker ComposeTraefik

Roadmap

CHAT

LLM endpoint — local or cloud

Single OpenAI-compatible REST surface. Swappable backend: Ollama for local inference, or a proxy to OpenAI / Anthropic / any compatible provider. Switch per workspace.

OllamaOpenAIAnthropic

Roadmap

RAG

Retrieval over your documents

Bring your own vector store. Local Chroma / Qdrant for fully on-prem deployments, or a remote managed one. Embeddings on CPU or via a provider. Plugs into CHAT when it lands.

ChromaQdrantmanaged

// Get Started

Install, verify, extend.

One command on Linux (Debian/Ubuntu, Fedora/RHEL, Arch) and macOS. Windows via WSL2. The installer detects existing Docker and reuses it — it does not reinstall or uninstall anything you have running. From there, install any module — local CPU, cloud-backed, your own.

Install the kernel

The installer is idempotent: rerunning it updates the kernel without touching your existing containers.

curl -fsSL https://queai.dev/install.sh | bash

Verify it's up

Health endpoint reports kernel version and active module count.

curl -fsSL http://localhost:8473/health

Open the dashboard

Sign in with the admin you created during install. From the catalog you install, configure and monitor modules with one click each.

open http://localhost:8473

Automate from the CLI

Optional. Install the queai CLI to drive the same operations from scripts, CI or another host through the REST API.

pipx install queai

QueAI Architecture

Browser / Dashboard UI

↕

QUEAI CORE

↕

● Modules (containers)

OCR — local CPU

STT — local CPU

TTS — local CPU

CHAT — local or cloud proxy

↕

Docker Engine

System Requirements

Dockerv24 or later

RAM8 GB min (16 GB recommended)

CPUx86_64 or arm64 (modules run on CPU)

OSLinux · macOS · Windows WSL2

InternetOnly for install & module downloads

// Open Source

MIT-licensed.
Built in public.

Source, roadmap and CI are public. The plugin contract is a single manifest.json plus a docker-compose.yml — whatever runs in a container can become a module: local models, cloud proxies, pipelines.

Publish a module: manifest.json + Dockerfile, the kernel does the rest.

Open issues live on GitHub. Newcomer-friendly ones carry the good first issue label.

Roadmap (Phase 0 → v1.0) tracked in docs/ROADMAP.md of the kernel repo.

GitHub

Code, issues and pull requests

→

Issues

Bug reports, feature requests, questions

→

Documentation

Architecture, operations, API reference

→