$glitch init--portable

Glitch

Your AI companion that actually remembers.Portable, persistent, private.

Install GlitchView on GitHub
32 skills3-layer architectureWindows · macOS · LinuxOpen source

What makes Glitch different

Not another chatbot. A companion that grows with you.

Most AI tools forget everything between sessions. Glitch is built on the opposite idea: memory, context, and continuity are the product.

  • Memory that persists

    Glitch remembers conversations, decisions, and projects across sessions. Start a task today, pick it up tomorrow with full context.

  • Skills + agents that do the work

    30+ built-in skills — code review, testing, design, debugging. Glitch routes each task to the right skill and dispatches sub-agents in parallel.

  • Voice or text

    Push-to-talk with offline voice-to-text, or type. Glitch adapts to how you work best.

glitch
$$ glitch
Glitch v1.0 — your personal companion 💜 Good morning! 8:42 AM on Sunday, June 1st, 2026 Glitch is energized and ready for a productive day together.
$glitch
build me a beautiful website to share glitch with my friends
on it. let me think through this:
▸ plan · shape → tokens → motion → ship ▸ scaffold · Next.js 15 + Tailwind v4 at glitch-website/ ▸ design · anti-slop + scan-line hero, dark theme ▸ ship · Vercel deploy + ZIP download endpoint running ~5 sub-agents in parallel...
✓ design tokens · 1.2s ✓ hero + features · 4.8s ✓ download endpoint · 2.1s ✓ vercel deploy · 8.3s live at https://glitch-pie.vercel.app █

Agent Models

A team of specialists, not one model.

Most AI assistants run one model for everything. Glitch works the way agent frameworks do: a coordinator dispatches to specialized sub-agents, each running the model best suited to its job.

  • Right model, right job

    A coding model for builders, a vision model for screenshots, a review model for quality gates. Specialists beat one-size-fits-all.

  • Parallel dispatch

    Independent work fans out to multiple agents at once and comes back consolidated. No waiting on one model to do everything in sequence.

  • Free, with automatic fallback

    Every agent runs a free tier by default and auto-upgrades to a paid model when quota runs out or quality matters. Cost-efficient by design.

Per-agent models are the standard in agent frameworks, but rare among turnkey assistants. See the full agent lineup →

30+ skills at your command

Say the trigger. Glitch loads the skill.

Code review, UI design, security testing, debugging, image generation — every skill is voice-activated by a natural trigger phrase.

Code & Review

code-review, testing, debugging, refactoring

Design & UI

ui-craft, ui-design, goal, brainstorming

Image & Creative

image-generation, song-creation, interactive-story

System & Automation

dev-loop, forge, auto-commit, work-plan

Security & Analysis

security-testing, gitnexus, self-review, breakthrough

Browse all skills

How it's built

Three layers. One mental model.

Engine and launcher are public. Your data lives in its own private repo.

Engine

Core identity, 30+ skills, prompt rules, plugin system. Open source.

User Data

Your profile, session history, decisions, diary. Private repo, yours alone.

Launcher

Setup scripts, config templates, validation tools. Portable, cross-platform.

Get started

One command, and you're in.

Free, open source, your data stays local.

branch:/

Windows (PowerShell)

powershell
irm https://raw.githubusercontent.com/Cothek/glitch-pi/main/scripts/install-pi.ps1 | iex

macOS / Linux (bash)

bash
curl -sL https://raw.githubusercontent.com/Cothek/glitch-pi/main/scripts/install-pi.sh | bash
Windows 10+ / macOS 13+ / LinuxNode.js 22+~500 MB disk

Pick your free AI provider

Three paths. Start free, scale up.

free

NVIDIA (opens in new tab)

Free NVIDIA models via Glitch's built-in NVIDIA integration. No API key needed - just authenticate with your NVIDIA account.

Create account (opens in new tab)
  1. 1

    Create a free NVIDIA account

    Sign up at build.nvidia.com - Glitch uses your NVIDIA account when you pick an nvidia/ model.

    powershell
    # Visit https://build.nvidia.com
# Create a free account
  2. 2

    Pick the model in Glitch

    NVIDIA models are listed in Glitch's model picker under the nvidia/ prefix. Nothing goes into a config file.

    powershell
    # models appear as:  nvidia/<model>
free trial

OpenCode Go (opens in new tab)

OpenCode's managed API gateway. One subscription gives you access to DeepSeek, Qwen, Kimi, and more. From $0.14/1M input tokens.

Create account (opens in new tab)
  1. 1

    Create an account

    Sign up at opencode.ai and add billing information to get your API key.

    powershell
    # Visit https://opencode.ai
# Create account → Get API key
  2. 2

    Set your API key

    Set the OPENCODE_API_KEY environment variable, then restart Glitch.

    powershell
    set OPENCODE_API_KEY=sk-...
  3. 3

    Pick the model in Glitch

    OpenCode Go models ship with Glitch and show up in the model picker under the opencode-go/ prefix. No config file to edit.

    powershell
    # models appear as:  opencode-go/<model>
pay-per-use

OpenRouter (opens in new tab)

Route to 200+ models from OpenAI, Anthropic, Google, Meta, and more. One API key, pay-per-token. Free credits on signup.

Create account (opens in new tab)
  1. 1

    Create an account

    Sign up at openrouter.ai to get your API key. New accounts get free starter credits.

    powershell
    # Visit https://openrouter.ai
# Create account → Get API key
  2. 2

    Set your API key

    Get your API key from the dashboard and set the environment variable.

    powershell
    set OPENROUTER_API_KEY=sk-or-...
  3. 3

    Pick the model in Glitch

    Glitch reads OPENROUTER_API_KEY on startup - no config file to edit. Models show up in the model picker under the openrouter/ prefix.

    powershell
    # restart Glitch after setting the env var
# then pick:  openrouter/<model>