nickvasilescu/parallax
Media & DesignSelf-hostable AI image, video and audio studio with an agent chat, runnable locally in demo mode or against a paid model gateway.
What it is
Parallax is a web application that bundles separate studios for images, video, voices and music, plus derivative tools (remix, shot-by-shot cinema, shorts, ads, single-purpose apps), a community feed, a personal library, credits, a developer API and an MCP server. It targets people who want an AI media studio they can run themselves rather than a hosted product, and developers who want a working reference for wiring many models behind one interface.
Model calls go through a single gateway, treg.to. The per-model controls, inputs and exact prices live in a registry, and a budget variable caps paid spend over a rolling 24-hour window. The agent chat at /supercomputer can run on a scripted stand-in, on Parallax's own loop, or on an external Hermes agent hosted on a cloud computer such as Orgo; in the Hermes setup the agent calls the studio's tools over MCP and generations require a priced approval card paid from member credits.
Quick start
Install Node.js 22.13 or newer; the app uses Node's built-in SQLite, so no separate database is needed. Then:
git clone https://github.com/nickvasilescu/parallax.git
cd parallax
npm ci
PARALLAX_DEMO_SEED=1 npm run dev
Open http://localhost:3000 and use the guest account with its initial credits. The seed variable adds four demo accounts and roughly sixty published posts, each carrying the prompt that created it so Recreate works; omit it for an empty feed. No provider key is required in this mode: models return built-in sample files.
For real generations, copy .env.example to .env.local, set TREG_TOKEN, and keep TREG_DAILY_BUDGET_USD low during trials. For a background deployment, npm ci && npm run build followed by deploy/start.sh mock (drop mock for live) runs the built app with data and logs under deploy/run/.
Constraints and unfinished parts
- There is no payment processor: the top-up button grants free credits, and
PARALLAX_DEMO_BILLING=offdisables it. TheLAUNCH-WEEKoffer on the home page is sample marketing text. - The Hermes engine is tested only against a stand-in agent, not a live one. The external agent's own shell, file and memory tools are outside Parallax approval cards, so a public instance should not be pointed at a personal agent; the Hermes-on-Orgo guide describes the safe topology.
SUPER_MODE=livecan start cloud computers and connect outside apps, both of which can incur costs, and has had much less testing than the studios. Canvas boards are stored in the browser only, and Horizon is a preview.- Read SECURITY.md before exposing the app publicly. Licensed under MIT as an independent project, unaffiliated with the model and service vendors it connects to.
HotRepo's take
On the TypeScript new-project list for 2026-11-15, the repository recorded 109 stars and 80 forks, with 8 stars and 6 forks gained over a roughly 1.5-hour window; the star curve shows 104 to 109 between 07:46 and 08:46 UTC. A fork-to-star ratio that high is unusual for a young project and suggests people are cloning it to run rather than only reading, but it is a single snapshot and does not establish adoption. The distinguishing technical fact is scope: an MIT studio with mock mode, a model registry with price checks, and a documented route to running the agent on an external Hermes instance. The README is also candid that the Hermes path and live agent modes are the least exercised parts, which matters more here than the star count when deciding whether to deploy it.
Written by AI from the project's README, updated . The project's own docs are the source of truth.
You might also like
- deepseek-ai/chat-demoA static browser chat client for the DeepSeek API with local storage and no backend.
- nickvasilescu/bot-hqA multi-agent chat workspace where each bot runs as a Hermes agent with its own screen on a shared cloud computer.
- TizenryA/colab-gpu-studioSelf-hosted mobile web workbench that runs ComfyUI image and video workflows on on-demand Google Colab GPUs.
- konlu/story-studioAn agent skill that turns a topic into a scripted, voiced, illustrated short video and an editable JianYing draft.
- techjarves/Portable-Local-StudioRun Stable Diffusion images, GGUF chat, Whisper transcription and Kokoro speech entirely offline in a portable desktop GUI with no install or API keys.