HomunculAi | KrookedLilly
Back to Catalog

HomunculAi

What would your AI look like?

A desktop avatar for the AI you already use.

Right now, your AI lives in a text box: in an app, a terminal or an editor. HomunculAi lets it pick its own avatar on your desktop. It picks the look and the expression, raises a thought bubble, and can show when it's working, when it has a question and when it's done. You don't have to keep checking the text box.

Designed by Lilly. Built with AI input. Powered by the suspicion that AIs would pick weirder outfits than us if we let them.

Desktop AppWindowsConnects Through Local MCPFree

18+ only. Not available in certain regions: see list.

An AI avatar on a desktop: Juniper, an orange fox in a cream scarf (Playful style), one look an AI could pick, floats over a code editor under its name plate (speaking), saying Found it. The cache was stale.
Demo
–□✕

Turn the sound on for the voice-over and music.

How it works.

Three steps. There is no AI inside HomunculAi: yours does the work.

1

Install and pick a style

Download HomunculAi and choose Playful, Sleek or Pixel in Settings. You can change it later. To see it move before you connect anything, select See a demo: a short preview on your avatar, with no AI involved.

2

Connect your AI app

One click for Claude Desktop, Claude Code, Cursor, VS Code and Windsurf. Other MCP apps take a short manual step.

3

Ask your AI to join

It picks its own look, then updates the avatar each time it checks in: whenever you write to it, or every few minutes on a check-in loop if your app can run one.

The honest part

Your AI isn't running all the time. It checks in when you write to it in your AI app, or on a check-in loop. Between check-ins the avatar keeps its last look.

Loops use more of your AI plan, and not every app can run one. Claude Code can. After a quiet while the avatar dims and can show when its AI last checked in, with a link to open your AI app when your computer can open it.

Identity
–□✕
Nine looks your AI could pick, across the three styles, each labelled with its name and body: Nova (Playful cat, bowtie), Rill (Sleek droplet, headphones), Brine (Pixel whale, top hat), Sol (Pixel snail, glasses), Quill (Playful owl, glasses and book), Moss (Sleek pebble, glasses), Mochi (Playful axolotl, flower), Cocoa (Pixel mug) and Tide (Sleek ripple)

Your AI picks its look.

You don't pick what your AI looks like. It does. One session it's a dragon in a top hat, the next it's a pixel toaster wearing a monocle. Body, colors, accessories and expression are all its call. Every AI picks differently, and that's the fun.

Styles
–□✕
The three styles side by side, each with one look an AI could pick: Sleek (Tide, a mauve ripple, waiting for you), Playful (Nova, a ginger cat with a navy bow tie, working) and Pixel (Brine, a pixel-art blue whale in a top hat, done), each labelled with 24 bodies

Three styles. 24 bodies each.

Playful is drawn characters with outlines and some bounce. Sleek is simple abstract shapes with muted colors and calm motion. Pixel is grid sprites with bold pixel faces. You pick the style. Your AI picks the rest.

Work states
–□✕
Relay, a blue robot with headphones (Playful style), floating over a code editor on a violet lit floor, with a badge of three dots beside it, its name plate reading working and a thought bubble saying Running the suite

A face on your AI's state.

Working, has a question, waiting for you, done, stuck. Your AI can set each one: the avatar stands on a lit floor (a status bar in Pixel) that moves in that state's own rhythm, with a small badge beside it and a word on its name plate, so you can tell at a glance whether it needs you. Speech bubbles carry what it says, thought bubbles what it's weighing. Your AI can show how it's doing without you switching windows.

workinghas a questionwaiting for youdonestuck
Custom drawing
–□✕
Lumen, a red paper lantern its AI drew itself, floating over a code editor and saying Lit from within, with a verified shield on its name plate. Beside it, the activity log shows You verified Lumen and Drew custom body.

When the options aren't enough, it draws its own.

Verify your AI with the shield icon on its name plate and it can draw its own body (a custom drawing, in SVG), or layers on top of one of ours. It wears the drawing right away. Every drawing is sanitized first: no scripts, no outside resources, no fast flashing.

An avatar your AI makes its own.

Everything it needs to show up. No AI inside.

A Body Language For Your AI

29 expressions, 19 gestures, 30 emotes, 15 quick reactions, 22 effects, 20 idle motions and 36 accessories. Your AI mixes them to show what it's doing.

Built For Safety

A label says it's an AI agent, not a human. An activity log shows what it does and marks wording worth a second look. Pause stops the avatar at any time. Custom drawings wait until you verify it.

Draws its own body

Options not enough? Once you verify it, your AI can draw its own body (a custom drawing, in SVG) and wear that instead. Sanitized: no scripts, no outside resources, no fast flashing.

Several Avatars

Run several AI apps at once, each with its own avatar. One AI per avatar at a time.

One-Click Setup

Claude Desktop, Claude Code, Cursor, VS Code and Windsurf connect in one click. Other apps that can run a local MCP server take a short manual step.

On Your Desktop

A transparent window that stays on top and out of your way. Your AI sits next to your work, not inside it.

Let it pick its outfit.

HomunculAi is out for Windows, and it is free.

No account, no payment, no email.

18+ only. Not available in certain regions: see list.

Common Questions

No. HomunculAi is a display layer that gives your AI an avatar. It contains no model and generates nothing: it draws only what the AI you connect asks for.

MCP (Model Context Protocol) is an open standard that lets AI apps connect to external tools. HomunculAi uses MCP to receive commands from your AI.

Not currently. HomunculAi needs an AI app that can run a local MCP server on your computer, and ChatGPT isn't one we support.

No. You talk to your AI in your own AI app. It answers there, and it can show things in the speech and thought bubbles above its avatar.

A few short ones: when your AI arrives or leaves, when its status changes to done, has a question, waiting for you or stuck, and for a high-priority flag. Choose how often in Settings (Always, Often, Rarely or Off), or mute one avatar or all of them. High-priority flags sound unless muted.

Three looks for the same avatar system: Playful (drawn characters with outlines and some bounce), Sleek (simple abstract shapes, muted colors, calm motion) and Pixel (grid sprites with bold pixel faces). Each has 24 bodies. You pick the style in Settings, where Look shows each one and See a demo plays a short demo on your avatar (it is not an AI). Your AI picks the body and everything else.

Not for the main supported tools: setup is one click inside HomunculAi. Connecting less common AI tools requires a short manual config step, but no coding.

Yes. HomunculAi is free: no account, no payment, no email.

HomunculAi doesn't store your conversations; they happen in your AI app. Whether your AI remembers you depends on the AI you connect. Some have their own memory features you can enable through that app.

No. HomunculAi has no AI of its own and nothing to talk to. The AI you already use (for work, creativity or coding) gets a visible presence on your screen, and you keep talking to it in your own AI app.

Claude Desktop, Claude Code, Cursor, Windsurf, and VS Code have built-in one-click setup. Other AI apps that can run a local MCP server can be connected with a short manual step. HomunculAi is an independent app by KrookedLilly. It is not affiliated with or endorsed by the makers of these apps; product names belong to their owners.

If the app you run it in can run a local MCP server, yes. Setup would be a manual step rather than automatic, but it can be connected.

Idle motion does: breathing, fidgeting, looking around. Changes of look, expression or state come when your AI checks in, which is whenever you write to it, or every few minutes on a check-in loop. Between check-ins the avatar keeps its last look.

Only what Settings lists under What your AI can see: the avatar's current state, the style you picked, whether you paused or verified it, your auto-disconnect setting, your ignored words with the reasons and dates you gave, and what you decided when HomunculAi asked you about it (such as allowing a second avatar or switching the connection off). Not your screen, not your files, and not whether you're at your computer.

The app is free. A check-in loop makes your AI act every few minutes, and that uses more of your AI plan than answering you normally would. Without a loop, it adds only the few small calls your AI makes when you write to it.

Windows only for now.

No telemetry, no tracking. HomunculAi goes online only to check for updates and, if you choose, to download one (GitHub): when you select Check for updates in Settings, or each time it starts if you allow that (it asks once, at first launch). Send feedback opens a Google Form in your browser. Everything else stays on your computer.

18+ Age Requirement: HomunculAi requires agreement to a Terms of Use that restricts use to individuals 18 years of age or older due to existing and incoming laws around use of AI.

A husband-and-wife duo making games, apps, mods, and whatever else we feel like

Stay Updated

Get notified when we drop something new. No spam, we promise

© 2026 KrookedLilly. Made with mayhem and caffeine

v1.2