Skip to content

Familiar · Powered by VulRunWG

Stay in the game.
Use more of your hardware.

A local AI companion that can put your integrated graphics to work while your dedicated GPU renders the game.

On a compatible dual-GPU laptop, Familiar runs its AI workload on the integrated GPU. Ask by voice or keyboard, draw on your own game notebooks, and receive an answer in a floating overlay.

In development for Windows. No public download yet.

Inside the laptop demonstrationWork allocation

One laptop. Different jobs.

Dedicated GPURTX 4070 Laptop

Game rendering

DirectX 12
Integrated GPURadeon 780M

Familiar inference

VulRunWG / Vulkan
Ryzen 9 8945HSCPU · Input, coordination and audio playback
Recorded laptop configuration. This diagram explains the roles; it is not a live utilization meter.

The hardware you already own

Another processor.
A useful job.

A laptop can have capable integrated graphics alongside its dedicated gaming GPU. Familiar can use that integrated processor for local AI, giving it work that would otherwise compete directly for the gaming GPU.

The recorded laptop demonstration uses a Radeon 780M for Familiar and an RTX 4070 Laptop GPU for game rendering. The CPU handles input, coordination and audio playback alongside its other work.

This is separate workloads on separate GPUs. It does not combine their memory or split one model across both cards. Shared system memory, power and cooling still matter; gaming impact depends on the setup.

From a question to an answer

A local pipeline you can understand.

Speech recognition, notebook retrieval, language generation and optional voice synthesis each have a different job. They work together through Familiar; the models run on your computer.

  1. Whisper

    Hear the question

    A local speech model turns the recording into text. Typed questions skip this step.

  2. Notebook retrieval

    Find relevant notes

    Relevant passages from the selected game notebook provide context for the answer.

  3. Local language model

    Generate a reply

    VulRunWG runs the supported model and streams its answer into the overlay.

  4. Optional SpeechT5

    Read it aloud

    With spoken replies enabled, local voice synthesis prepares audio for playback.

What affects the time to an answer?

Recording transcription, finding context, generating text and preparing the voice all contribute. Model loading and first-use shader work can make a cold request slower. An answer already available from recall is different from a newly generated response. End-to-end timing should show these stages rather than presenting token speed as the whole experience.

Built around the player

Your notes. Within reach.

A notebook for each game

Import and export .familiar notebooks containing your guides and notes. Selected reference passages help ground answers in the material you supplied. You can edit and share the notebook.

Keyboard or voice

Ctrl + Alt + Space summons the overlay. F9 is the default push-to-talk key. Text replies are the default, with local spoken replies available when enabled and the required models are installed.

Your own companion

Familiar is a separate application powered by VulRunWG. Its focus is a small overlay, game notebooks and quick access while playing. VulRunWG: Suite is the broader everyday AI workspace.

Know the fit

Local models.
Real requirements.

The integrated-GPU configuration requires a capable iGPU, a compatible driver and enough memory for the selected models. Hardware detection chooses a supported device path; not every computer has a suitable second GPU.

Use a windowed or borderless game for the overlay. Exclusive fullscreen may hide it. Check hotkeys and performance on your setup.

Notebooks are reference material, not live access to your character, quest progress or screen. Answers depend on the model and the accuracy of your notes. Model-publisher licenses remain applicable.

See how VulRunWG does the compute