Pinokio

The people who get the most from Pinokio are those who want to run models on hardware they control instead of paying per call to a hosted endpoint. The Pinokio store spans video generation such as Mini Max H3 in Comfy UI and Maestro, music and audio tools including Song Generation Studio, Stable DAW and Underfit for Stable Audio 3 LoRA fine-tuning, speech work through Drama Box TTS with voice cloning, 3D via Tripo Splat, local agents like Odysseus and Agents View, and utilities such as the SearXNG metasearch engine and the Geo Libre GIS viewer.

A typical session with Pinokio starts in the store, which filters by type (app, plugin, api), by platform (macOS, Windows, Linux), and by GPU family (NVIDIA, AMD, Apple), then sorts by Recommended, Latest, or Check-ins. Every Pinokio listing exposes the repository path, a version string such as v8.0, an updated-on date, tags like #video-generation or #tts, the publisher handle, and a check-in count that other users contributed after a successful install. Once installed, the application stays in the local Pinokio library so relaunching is one more click.

Pinokio also functions as a release feed. The Launcher updates section on pinokio.co surfaces posts from publishers including @bizarro, @blizaine, @mvrdvk, @iaplay, and @estebandezafra, so a Pinokio user sees when Phosphene ships a video editor or Maestro reaches v1.9.0 with a generation queue without subscribing to separate repositories. Pinokio was also advertising a beta tester call for an upcoming major feature update at the time of review.

On packaging, pinokio.co publishes no pricing page, no plan table, and no checkout, so Pinokio is presented as free software with a community-maintained script catalog. Because Pinokio also publishes no about page, terms page, or llms.txt on the public site, anyone assessing Pinokio for commercial, team, or regulated deployments will find licensing terms, support commitments, and any future monetization undocumented rather than merely cheap.

The true cost of Pinokio is hardware and disk space. Several headline Pinokio entries are hardware-locked: Mini Max H3 Comfy UI and Song Generation Studio are marked NVIDIA only, Wan2GP AMD is marked AMD only, ds4-webui is a Metal-only DeepSeek V4 Flash launcher, and Maestro requires an NVIDIA GPU with 6GB or more of VRAM. Even the disk-optimized Mini Max H3 build inside Pinokio lists roughly 63GB of pruned INT8/NVFP4 weights instead of about 290GB, so a handful of video models can eat hundreds of gigabytes.

Quality across Pinokio varies by publisher because the catalog is community driven rather than vendor curated. A Pinokio script with 140 check-ins such as Maestro carries far more social proof than a listing showing zero check-ins, and Pinokio does not claim to audit third-party install scripts. Running an entry from Pinokio means running community code with filesystem and network access, which matters on shared or managed machines.

Measured against hosted inference gateways and model APIs, Pinokio trades elastic capacity, uptime guarantees, and managed billing for privacy, no per-token cost, and complete control over checkpoints. Measured against single-purpose local runtimes that concentrate on chat models, Pinokio is deliberately broad, covering diffusion video pipelines, DAW-style audio production, transcription, GIS, search, and agent workspaces under one launcher.