Pinokio
Pinokio is a free desktop application for macOS, Windows and Linux that installs, launches and controls open-source AI apps entirely on your own machine, with every runtime bundled in and any AI agent able to drive them.
What is Pinokio?
Pinokio is a free desktop application for macOS, Windows and Linux whose job is to remove the setup work standing between you and an open-source AI project. Its own documentation calls it the Localhost Cloud: a cloud that runs privately on your own machines, or, in the manual's own comparison, Vercel, but for localhost.
Concretely, Pinokio ships with the toolchain those projects expect — Python, Node.js, Bun, Git, uv, Conda, Homebrew, FFmpeg, Caddy, and tmux on macOS and Linux — and can install anything else programmatically. You pick an app from the community catalogue at the heart of the project, click once, and Pinokio clones the repository, builds an isolated environment and installs the dependencies inside the app's own folder. Nothing is asked of you at the command line.
That catalogue is filterable by platform and by GPU vendor, and it leans heavily towards generative work: on 16 August 2026 it carried 187 entries tagged ai, 56 tagged tts, 44 image-generation, 38 video-generation, 24 agent and 21 music, spread over nine pages for a single listing path.
Around that core sit the features that make a personal machine behave like a hosted service. Every installed app receives a stable HTTPS address of the form comfyui.localhost, and any other server already running locally — Ollama, LM Studio, ComfyUI Desktop — picks one up as well. Apps can start automatically with Pinokio and declare dependencies on each other, resolved recursively. A one-click version control layer commits changes, switches between versions and publishes repositories to GitHub.
The most distinctive layer is the agent one. Pinokio writes two skill files, pinokio and gepeto, into the standard agent skills folder, where Codex CLI, Codex Desktop, Gemini CLI and comparable tools find them unaided. From there an agent can search your installed apps, start one that is not running, wait until it is ready and call it, with no MCP server and no skill file written by hand. Everything passing through a terminal is logged to flat files, which doubles as a shared memory any agent can read.
What it does
- Install an open-source AI app in one click from the community catalogue
- Launch and stop local servers without touching a terminal
- Let any AI agent discover, start and operate the apps you have installed
- Give every local app a fixed HTTPS address instead of a shifting port number
- Run several copies or versions of the same app side by side
- Package your own one-click launcher and publish it to GitHub
- Browse every server running on your machine and across your local network
When to use Pinokio / When not to
A quick filter to help you decide if Pinokio is the right fit.
When to use Pinokio
- Curious users who want to try open-source AI apps without setting up Python, CUDA or virtual environments by hand
- Creatives generating images, video, music or speech locally rather than through a paid cloud service
- Developers and researchers who need a reproducible local environment with the toolchain already in place
- People already working with AI agents such as Claude Code, Codex CLI or Gemini CLI and wanting them to drive local apps
- Privacy-conscious users who would rather keep prompts, files and models on their own hardware
When not to use Pinokio
- Anyone without a capable machine: many catalogue apps demand an NVIDIA GPU, sometimes 6 GB of VRAM or more
- Teams looking for a hosted SaaS, since there is no cloud version and everything runs on your own hardware
- Mobile-first users, as Pinokio ships only as a desktop application with no iOS or Android app
- Buyers who need a data processing agreement, security certifications or a named legal entity before signing anything
- Users expecting a vetted, guaranteed app store: the catalogue is community-run and the terms disclaim any warranty on it
How to use Pinokio
A typical end-to-end flow, from setup to results.
- Download the build for your platform: macOS on Apple Silicon or Intel, a 64-bit Windows installer, or Linux as AppImage, DEB or RPM in x64 or ARM64
- Install and open the application; the runtimes ship with it, so nothing else needs installing first
- Browse the app catalogue and narrow it down with the platform and GPU filters before choosing
- Install an app in one click and let Pinokio clone the repository and build its isolated environment
- Launch the app; it opens inside Pinokio and receives a fixed local HTTPS address
- Optionally enable auto-launch at startup and declare which apps must be ready before others start
- Open any AI agent: the pinokio and gepeto skills are already in place and trigger on ordinary requests, such as generating a speech clip
- For agents that do not scan the standard skills folder, download the skill files from the Settings page and import them by hand
- Build your own launcher from the Create button, let an AI agent structure the project, then publish it to GitHub in one click
- To be featured on the Discover page, get verified by the maintainer, join the Pinokio Factory GitHub organisation, transfer your repository and request a review
Pros & Cons
Pros
- Free and MIT-licensed, with no account required to use the desktop application
- Removes the environment setup that stops most people from ever trying open-source AI apps
- Everything runs locally, so working files and prompts never reach a third-party service
- Covers macOS, Windows and Linux, down to ARM64 builds
- AI agents can drive installed apps with no MCP server and no skill file to write
- Package freshness delays are enabled by default, limiting exposure to recently compromised releases
- Scripts are readable JSON, run isolated under a single folder, and featured ones are manually reviewed
Cons
- Usefulness depends on your hardware: many catalogue apps require an NVIDIA GPU and a stated amount of VRAM
- The terms disclaim any guarantee that indexed app metadata is complete, current, safe, accurate or endorsed
- Only apps featured on the Discover page go through a review; the rest do not
- The install freshness delay does not cover npx, uvx, plain pip, conda, brew or cargo
- No legal entity, postal address or support email address is published anywhere on the site
- No security certification, no data processing agreement and no mention of the GDPR
- No mobile app, and the documentation is a JavaScript-rendered page that shows nothing without scripts
Pricing & Plans
Pinokio is free of charge. The desktop application is distributed under the MIT licence, and the terms of service state, under a heading titled No Paid Services, that the Registry includes no payments, no subscriptions and no paid plans. There is consequently no pricing page, no entry price point and no currency to report. The terms reserve the possibility of introducing paid features later, under additional payment terms. The genuine cost falls on the user's own equipment: a suitable graphics card and the disk space taken by downloaded models.
- Free — full access to the desktop application under the MIT licence
- no account required
- Free Registry account — optional
- adds posting
- commenting
- following
- saving
- check-ins and app submissions
Data, GDPR & hosting
A consolidated view of how Pinokio handles your data.
GDPR overview
There is no GDPR implementation to report. The words GDPR and General Data Protection Regulation appear nowhere on the site — not in the privacy policy dated 4 June 2026, not in the terms of service, not in any other page. No lawful basis is cited, no Article 27 EU representative is designated, no data protection officer is named, and neither standard contractual clauses nor a data processing agreement are offered. Some rights do exist in practice, but framed as product settings rather than legal entitlements: editing your profile and notification preferences, unlinking your GitHub account, controlling the visibility of saves and check-ins, deleting your own posts and comments, and requesting access, correction or deletion by writing to privacy@pinokio.co. The policy also warns that information may be processed in the United States and other countries whose protection laws differ from your own.
Who owns the data?
Ownership splits in two. Apps installed through Pinokio run on your own machine, so the files and models they produce never leave it. On the Pinokio Registry, the online directory, the terms state that you keep ownership of everything you submit — posts, comments, profile text, tags, uploaded media, check-ins and app submissions — while granting the operators a non-exclusive, worldwide, royalty-free licence to host, store, reproduce, reformat, display, distribute and use that content to run, secure, improve and promote the service. The Registry claims no ownership over the third-party repositories, models, scripts or applications listed on its app pages, and repository owners keep their rights and their open-source licences.
Reuse rights
For content you post yourself, reuse is unrestricted: you keep ownership, and the licence you grant runs to the operators rather than away from you. Everything else the Registry displays stays governed by the original repository's own open-source licence, which the terms explicitly leave untouched — Pinokio indexes public metadata, it does not relicense it. Before installing or running any third-party script, model or package, the terms put the burden squarely on you to review its source, licence, permissions, network behaviour and system requirements. Anything you publish on the Registry is public by design and may be seen by other users and indexed by search engines.
Data retention & training
Hosting summary
Two very different pictures sit behind the same name. Apps installed through Pinokio run on the user's own machine: models, files and logs stay in the local Pinokio folder and in each project directory, under the user's sole control. The online Pinokio Registry is a hosted service, and its privacy policy says only that it may be operated from, and information may be processed in, the United States and other countries where the operators or their service providers work, warning that those countries may have different data protection laws. No specific hosting country, region or data centre is named. No subprocessor list is published either; providers are described by category — hosting, security, storage, delivery, email, analytics and debugging. Three third parties surface indirectly: Cloudflare Turnstile for security checks, GitHub for OAuth and repository metadata, and Hugging Face or X for fetching public content from submitted URLs. No SOC 2, ISO 27001 or comparable certification is claimed anywhere. On the network side, the domain is registered with Squarespace Domains II LLC and resolves to an address located in the United States.
Things to keep in mind
Risks and trade-offs to weigh before adopting Pinokio.
- Installing third-party code is the core activity here, and the terms place the whole burden of reviewing source, licence, permissions and network behaviour on you
- Only featured apps are reviewed, yet a one-click install still executes someone else's script on your machine
- The freshness delay guarding against newly compromised packages does not apply to npx, uvx, plain pip, conda, brew or cargo
- Anything posted to the Registry is public and indexable, and moderated or deleted content can survive in backups
- There is no identifiable publisher to hold to account: no company name, no address, no support contact, and liability capped at 100 US dollars
- Local AI is easy to over-consume: models fill disks, long generations burn hours, and the absence of a bill removes the usual brake
- The project's original domain now redirects elsewhere, so check that any download really comes from the official desktop site and not a lookalike
Setup & Integrations
Technical difficulty
Low for the software itself, variable for what you run on it. Installation is an ordinary platform installer, and the download page suggests the right macOS build automatically. No runtime needs installing separately, since Python, Node.js, Bun, Git, uv, Conda, Homebrew, FFmpeg and Caddy ship inside. Installing an app is one click. The real difficulty is hardware: picking a catalogue entry your GPU can actually run, with the VRAM it asks for. Building your own launcher or plugin is a further step, though the documented JSON scripting API and the built-in gepeto agent do most of the work.
Deployment
Integrations
Behind Pinokio
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
Does Pinokio cost anything?
Which operating systems does it support?
Do I need a graphics card?
Does my data go to the cloud?
Do I need an account?
Is there a minimum age?
Are the apps in the catalogue vetted?
How do AI agents control my apps?
How do I get support?
Can I publish my own launcher?
Should you pick Pinokio?
Pinokio solves a narrow, real and widely felt problem: the environment setup standing between a promising open-source AI project and actually running it. It is free, MIT-licensed, asks for no account and no commitment, and it does the unglamorous work — cloning, isolating, installing, launching — well enough that a catalogue entry genuinely becomes a running app.
Its most interesting move is the agent layer. Dropping two skill files into a standard folder so that Codex CLI, Gemini CLI or Claude Code can search, start and call local apps without a line of glue code is a considerable convenience, and the flat-file logging behind it is refreshingly unclever: no database, no API, just files any tool can read.
The limits are equally clear. What you can actually run is decided by your graphics card, not by Pinokio. The catalogue is community-built, and the terms of service disclaim in plain words any guarantee that indexed metadata is complete, current, safe, accurate or endorsed by the repository owner; only apps featured on the Discover page pass a review. The default package freshness delay is a thoughtful safeguard, but it stops short of npx, uvx, plain pip, conda, brew and cargo.
Governance is the weakest point. No legal entity, postal address, support email or named individual appears anywhere on the site, and the legal documents cover the online Registry rather than the desktop software. There is no data processing agreement, no certification and no mention of the GDPR.
For personal, hobby or laboratory use, Pinokio is an excellent entry point into local AI and hard to beat at its price. For procurement that needs an identified counterparty and audited controls, the paperwork simply is not there.
- Choosing a selection results in a full page refresh.
- Opens in a new window.