GLaDOS

GLaDOS is an open-source project building a real-life version of the iconic AI from Valve’s Portal series: a fully local voice assistant with GLaDOS’s custom voice, dry humour, camera vision, persistent memory and tools. It has earned about 5,750 stars on GitHub and is actively developed, with its latest updates in 2026.

Overview

Instead of a single monolithic chatbot, GLaDOS is organized into independent “cores” — separate modules for emotion, memory, vision, system health and research — whose observations feed a shared context, while an Autonomy Core decides when something deserves a spoken response. Conversation runs on local models: the reference setup serves Gemma 4 through llama.cpp, with speech handled by local ONNX models, so the assistant can run privately on your own hardware with CUDA, ROCm or CPU backends. A functional web console (plus a terminal UI) lets you control the microphone, voice and camera, inspect each core’s state and memory, and watch the animated GLaDOS eye react as it talks. Through MCP integrations it can connect to services such as Home Assistant and run allowlisted system commands.

Key Features

• Real-time voice conversation with a faithful GLaDOS voice and personality
• Runs fully locally: llama.cpp/Gemma conversation model plus ONNX speech models
• Vision core that observes camera scenes and describes changes
• Persistent memory: conversation summaries and facts are stored and retrieved
• Multi-core architecture — emotion, memory, vision, health, research and autonomy cores
• MCP tool integrations, including Home Assistant and native task/memory controls
• Web console and TUI to inspect minds, context and background tasks

💳 Pricing

GLaDOS is completely free and open source under the MIT license. You supply the hardware and any optional external models or services you configure.

Who Is It For

GLaDOS is for makers, homelab builders and Portal fans who want a private, always-on voice assistant with genuine personality — and for developers interested in multi-agent “core” architectures for embodied assistants.