Aether

An AI-driven OSINT analyst that lives on your desktop.

Give her a name, an email, a username, a domain, or a photo. She opens a case, runs the target across the open web, reads the metadata, maps the infrastructure, and draws everything she finds into a knowledge graph that grows while you watch. Runs on macOS and Windows, powered by Claude, ChatGPT, Gemini, or a local model of your own.

 

please don't use this to dox the innocent >:(

Seriously. Aether is for people and systems you're actually allowed to look into: your own exposure, folks who asked you to check theirs, and lab or CTF boxes you own. She only reads what's already public or shown by a platform. She won't break authentication, get past bot-detection, phish, or take over accounts, and those limits are baked into how she works. Point her at a stranger you have no business investigating and you're the baddie, not her. Be normal.

What it's like to use

You hand Aether a selector and she gets to work, writing up what she finds rather than firing off chat replies. Every tool she runs shows up as a numbered line in an evidence log that stays with the turn — so a finding has an address you can cite later, like 02.1. The moment she finds something she writes it into the graph, then flips its status as she confirms it or rules it out.

The graph encodes meaning rather than decorating with it. Shape is the selector type — circles for identity, squares for infrastructure, diamonds for artifacts, triangles for contact details. The ring is the status: a solid ring is confirmed, a ring broken at twelve o'clock is an open lead, dashed is a candidate, and an excluded node is hatched and struck through. Roman type is established, italic is provisional. All of it survives a greyscale screenshot, and all of it reads the same in either theme. Nodes carry real pictures too: a face on a person, the site's favicon on an account, the photo itself on a photo node.

Every turn is numbered and every tool call sub-numbered beneath it. Your words are kept verbatim in mono; her report is set in sans against a margin rule.

Selecting a node brackets it, draws a crosshair through it, lights its neighbourhood and drops everything else back a value step.

A first-class light theme, not an inverted dark one. The canvas is drawn from the same tokens as the rest of the app, so it repaints instantly when the theme changes.

At the default access level, Aether asks — whichever model is driving. The exact command is shown in full, never truncated, because a decision made on an ellipsis is not an informed one. The level itself sits in the chat next to the model; Shift+Tab toggles Safe and Ask.

One list, grouped by the question each capability answers. The switch says you want it — if it needs a program installed, turning it on installs that first.

First run asks one thing: which model. Claude, ChatGPT, Gemini or a local Ollama — set up whichever you pick right there, and the screen steps aside the moment that backend is reachable.

Run her on Claude, on ChatGPT, on Gemini, or fully local through Ollama. Same tools, same graph, same access rules, your choice of brain. Where a provider can list its models, the picker is that list.

What's in the box

A live knowledge graph. A force-directed canvas you can pan, zoom and drag. Marks are shaped by selector type, ringed by status, sized by how connected they are, and carry real pictures where there are any. A node written mid-turn pulses one expanding ring, so you can see which of the four things a tool just reported actually landed. It's the main workspace, and it updates as the case builds.

A transcript you can cite. Answers stream in as she writes them, and every tool call is a numbered line in an evidence log that is stored with the turn instead of vanishing when it finishes.

Light and dark, both first-class. One palette, defined twice, with every pairing contrast-checked. The graph canvas reads from the same tokens as the DOM, so switching theme repaints it with no reload.