The Open Arsenal.
Free tools that will run your lab. Each one explained in plain words: what it does, and how you start it.
Yerevan · 10-12 August 2026
0
Licences, subscriptions, vendors. The whole arsenal is free.
- No one can switch it off.
- Nothing leaves your room.
- The money goes to your analysts.
Every tool here is one of four kinds.
You open it in the browser and use it. Nothing to install.
Installed once in Chrome or Firefox, then it works while you browse.
Installed on the lab computer. We install everything together on Day 2; after that it just runs.
A downloadable brain for a specific job. It sits inside one simple app and works offline.
Stop scrolling. Start collecting.
What it does: Collects posts from Telegram and other platforms into one searchable table. Instead of reading a channel post by post, you ask questions of thousands of posts at once: what appeared when, from whom, how often.
How you start: It has a normal web-page interface once installed. We install it for your lab together on Day 2.
What it does: You scroll TikTok, Instagram or X the way you always do, and it quietly saves every post that passes your screen into a file you can analyse later in 4CAT.
How you start: Install once in the browser, press record, scroll. That is all.
Is this recording real?
What it does: Reads the hidden information inside a photo or video file: when it was really created, what software touched it, sometimes even where. The first question you ask of every suspicious file.
How you start: Drop a file onto it, read the answer. We set it up on the lab computer on Day 2.
What it does: You give it a voice recording, it tells you how likely the voice is computer-generated. The same kind of check that flagged the fake recording of the Prime Minister as 94% synthetic.
How you start: Lives inside the AI toolbox we install on Day 2: choose the file, press one button.
Every recording becomes text. In Armenian too.
What it does: Turns any audio or video into written text: a propaganda clip, a voice message, a two-hour stream. It understands Armenian. Suddenly everything is searchable and quotable.
How you start: Drop in an audio file, get text back. Works offline on the lab computer.
What it does: Translates between 200 languages, Armenian included. A Russian channel post, a Azerbaijani article, a French fake: readable in seconds, without sending the text to any website.
How you start: Paste text, pick the language. Part of the same offline AI toolbox.
10,000 posts in. 12 stories out.
What it does: Reads a huge pile of posts and groups the ones that tell the same story, even across languages. Ten thousand posts collapse into a dozen storylines you can actually look at.
How you start: On Day 2 we press the button together and you watch your own material sort itself.
What it does: You give it your list of known narratives ('you will end up like Ukraine', 'the PM sold the nation'...) and it tags every new post with the matching one. Your monitoring runs itself.
How you start: Same toolbox: define your list once, then it labels everything that comes in.
Who pushes this, together with whom?
What it does: Draws the picture: accounts as dots, connections as lines. Coordinated groups jump out visually. The network maps you saw in the reports in Part 1 are made with this.
How you start: A normal desktop program. Download, open, load your data.
What it does: You give it one suspicious website address. It comes back with everything public about it: who registered it, when, what other sites live next to it. This is how fake-site networks get unmasked.
How you start: Web-page interface after install; type an address, press start. Day 2.
What it does: One suspicious username, checked across hundreds of platforms in about a minute: where else does this account name exist? Great first move on an anonymous amplifier.
How you start: One typed command. We will wrap it in a clickable shortcut for the lab.
Watch a network appear.
From a table of accounts to a living map in minutes. This is the same walkthrough your analysts will follow on Day 2, on Armenian data.
Fakes get deleted. Your copy survives.
What it does: Your own permanent archive. Paste a link and it saves the whole page: text, images, video, with date and time. When the fake disappears next week, your evidence does not.
How you start: Runs on the lab computer with a simple web page: paste link, done.
What it does: Built by Bellingcat's investigators: you keep a simple spreadsheet of links, it archives every one of them automatically, the way courts and researchers need it.
How you start: Connects to a Google Sheet. We configure it once; then you only add links.
Like ChatGPT. Except private, free, and yours.
What it does: The app that runs AI models on your own computer. You chat with it like with ChatGPT, but nothing you type ever leaves the room. For an institute handling sensitive material, this is the whole point.
How you start: Install it, type one line: ollama run qwen3. That is genuinely all.
What it does: The AI brain we use daily in our own work: it reads piles of posts, summarises them, suggests which manipulation technique a post uses, drafts report sections. Speaks many languages.
How you start: Downloaded once through Ollama. After that: no internet needed, no fees, ever.
From install to first chat in one coffee.
A 15-minute walkthrough of everything on the previous slide: install Ollama, download a model, start asking questions. No account, no card, no cloud.
We use all of this every day.
30B
AI parameters running locally in Gliwice
0
cloud calls for sensitive material
1
suitcase: our working kit, not a demo
One shared dictionary for manipulation.
DISARM is a public, free dictionary of manipulation techniques: every trick has a name and a code. When your report says 'technique T0023', an analyst in Brussels knows exactly what you mean.
It does not exist in Armenian yet. Translating the core of it could be this project's visible gift to the field, with AIISA's name on it.
You will see all of it run.
- Whisper turns a propaganda clip into text.
- The grouper sorts headlines; you name the storylines.
- Local AI tags a post; a human says yes or no.
- The archive saves it all before we close the lid.
A lab you own outright.
- One good computer is enough.
- We set it up together in the 54-hour course.
- It stays yours. Nothing expires with the project.
Take it. All of it.
The machine attacking Armenia is industrial. The defence can be too, and it is free.