IllAIra CLM

What the app does

In order of use, from where you start to what you get. The same things on the Android app and in the WebApp, except where noted.

1. 1

   Take a Reminder from the Bazar

   You do

   Open the Bazar inside the app and add a free module to your profile, in Italian or English.

   You get

   An .md file of your own, already in the shape the model reads: identity, modules, room for memory.

2. 2

   Write as you speak, the app keeps the shape

   You do

   Open the file in the editor, in Markdown or in natural view, and write.

   You get

   The IllAIra standard stays in place without you having to know it. If the file changed underneath, the app tells you before overwriting.

3. 3

   See it as a Sphere Grid

   You do

   Tap the dot next to the file.

   You get

   The map of modules, sections and logs, with a pulse on what is switched on. Tap a node and edit from there.

4. 4

   Archive a log with one tap, without writing it by hand

   You do

   Ask the model for the session summary (IllAIra Logger, free in the Bazar) and copy it. On the App you tap MemoKen — the tile that stays on the edge of the screen even while you are in the chat, MEMO face — and the shelf proposes it; flip it to DROP and it hands the profile’s files to Gemini or ChatGPT in batches, through the Share sheet. In the WebApp you paste into «Add log», or you install the browser MemoKen, which brings it in with one click — and the shelf proposes it there too: from the names the AI wrote in the log, and by meaning if you have downloaded a model.

   You get

   The app puts the date and the header, and the entry lands at the bottom of the right topic’s zone. You always confirm: nothing writes itself. On the phone MemoKen is Android-App-only, it works in Manual mode and you turn it on from Settings — it starts off; in the browser it is an extension to install, and it stays capture. And the same log is not filed twice: if the entry is already in the module you chose — same date, same title, same text — the app tells you instead of appending it again, on the Android App and in the WebApp alike. With a downloaded model it also notices the same fact written in other words: “Looks like:” with the old entry, and you decide — add anyway, or open that one. Under every proposed destination, one line says why it is there.

5. 5

   Search by meaning, on the device

   You do

   Pick a model from Settings → Search by meaning: the app downloads it once and builds the index of your files. On the App there are four sizes (124, 347 and two of 612 MB); in the WebApp the first three — the largest runs on Android only, and the model card says so.

   You get

   You search “how I feel when…” and find the passage even if it doesn’t contain those words. All on the device: your texts never go online. Without a model, text search stays, and it is always there.

6. 6

   See the whole archive at once: the 3D Sphere Grid

   You do

   Open the Sphere Grid with the dot next to the file, then tap “3D” at the top. Long-pressing a dot offers the same entry, and from there the view opens with that node lit and brought to the front. It belongs to the Android App, and it asks for two things you already have if you search by meaning: a downloaded model and an indexed profile. If either is missing the screen tells you and takes you where it is fixed.

   You get

   The whole profile in three dimensions, like a body in shells: the profile at the centre, the files around it, the blocks by depth. The direction comes from the meaning of the text, so things that talk about the same matter end up in the same wedge and the lobes form by themselves. Closeness says similarity, not a link: the lines are only the ones the file declares.

7. 7

   See the links instead of the order: the “Understanding” screen

   You do

   From the Dashboard open «Understanding» — in Simplified mode it sits behind «⋯ › More options ›», from Intermediate it is in view — on the Android App and in the WebApp alike. The first level of links asks for nothing and is there from the first log you file with MemoKen: where the log went, and where the app had proposed putting it. For the second — entities and relations — you download from Settings → «Knowledge graph» a model that runs on the device, phone or browser. One section, and what you choose there is a depth, not a model: two cards with a dot, «Entities» and «Entities and relations», each saying in one line what it puts on the map and what it costs. Inside each card a single model is in view — the one in use, or the one recommended for your device, with its weight on the button — and the other sizes sit under «Other models». Under «Entities» there are four sizes — 201, 331, 379 or 647 MB; under «Entities and relations» two entries, and neither is a bigger size: 325 MB and, since 0.1.74, 411 MB, and they also find the relations between entities — that is, the lines of the cloud themselves. Whichever you pick, a single model is marked active, and the map reads that one. The first declares twenty-two languages, the second six — Italian, English, Spanish, French, German, Portuguese — and stops there. The second level also asks that the memory precision not be «by name»: at that level the map does not show what a model found, and says so. Relations are viewed in the «Relations» map of the same screen, and there are three views: 2D, 2.5D (the default) and 3D. In the WebApp the same six models run in the browser, from the same Settings: the browser files weigh a little more — 202, 332, 382, 650, 371 and 456 MB — and a model downloaded on one side is not shared with the other.

   You get

   A cloud where position says nothing — the dots sit where they settled — and it is the lines that speak: where you filed a log, where the app had suggested you file it, which block names which thing. These links are extracted, not guessed. With the model, every thing found — people, places, dates — becomes a dot that every block naming it attaches to; the 379 MB size is trained on Italian, English, Spanish, French and German, the other three declare twenty-two — Japanese, Korean, Bengali and Russian stay out of all four, and there it finds less. The four sizes find the things, not the relations between the things: people, places, dates. The relations — “lives in”, “owns” — are found by the two entries after the sizes, 325 MB and 411 MB, which are not bigger sizes but another capability: from there on a line can come from what a memory says, and not only from where you filed it. The 325 MB one is also the only one that declares Japanese, Korean, Bengali and Russian, but it has them from the engine it is built on, not from training on relations: how much it really finds there, nobody has measured. The 411 MB one, another engine of the same family, declares six languages — Italian, English, Spanish, French, German, Portuguese — and stops there. Both look at a 64-word window at a time: two things far apart in the same block can escape them. The third level is questions from the app, dashed lines you switch on yourself: they are not facts about you and they are not saved anywhere. None of this leaves the phone, or the browser.

8. 8

   Graft a module, merge two Reminders — or two modules

   You do

   Take a module from the Bazar, or a second Reminder, and merge it with yours. Two functional modules also merge with each other — Logger with Neural Tracing, a custom of yours with one taken from outside: in the WebApp and in the App.

   You get

   What is missing gets grafted in, what you already have is left alone, none of your lines are lost. New modules arrive switched off: you turn them on. A Reminder and a functional module do not merge with each other: they are two distinct gestures, with two lists that never meet.

9. 9

   Export and inject, into any model

   You do

   One tap on Export: the whole file, or a tree of chosen modules. On the Android App the DROP face of MemoKen hands the same files to a quick chat, in batches, through the Share sheet — not to a Project’s sources.

   You get

   On the phone, one tap of MemoKen and the files arrive in the right chat in batches of N, and it waits for all of them. On a computer, text ready to go into Claude, GPT, Gemini, Llama. The model gives language and compute; the file gives the who.

10. 10

    Talk with your AI in the WebApp, with the profile in front of it

    You do

    In the WebApp, open «Chat» from the Dashboard and choose who answers: your own key (DeepSeek, OpenRouter, OpenAI, Mistral, Anthropic), or a model on your computer (Ollama, LM Studio). Choose «Focus on my files» and the answer comes from the passages of the profile that match your message, with the Reminder in front. On the Android App the way is the MemoKen with your own AI.

    You get

    A reply that knows your Reminder and cites your modules, with the tokens under it and «Stop» halfway. Nothing is written into your files unless you save it with the MemoKen, and you confirm. Conversations are not memory: they stay on the device, per profile, and never sync. The key goes only to the provider you chose — IllAIra never sees it — and it disappears when you sign out.

11. 11

    Profiles, backups, and the twin in the browser

    You do

    One profile per identity; on the desktop open the WebApp with the same account.

    You get

    Same files. For a synced copy you can switch on our cloud for a profile — it stays off unless you choose it — or keep the folder in the cloud you already use: the files stay yours.

The app has three modes: Simplified shows the essential commands, Intermediate adds the tools to curate what you have — Understanding, the Markdown view, the map controls — and Advanced shows everything. Nothing is lost — you switch from Settings → Mode, and what a mode does not show sits under «More options ›». The WebApp has all three: in Simplified it names things by what they do for you — «Your AI», «Save this memory» — with the technical name within reach, in Intermediate by the technical name with the line that explains it. Nothing is lost: you switch from every screen and from Settings → Mode, on the App.

It is a beta: what entered in each version is on the what’s-new page. all the news, of the app and of the site

Latest published version: 0.1.159. all the news, from the app and the site