iSNEPS
It reads your documents, your inbox and your photos, remembers all of it, and acts on it — reminders, email, your home, the web. Use it on our servers and it works today. Run it on your own hardware, with your own keys, and nothing leaves the building. Same platform either way.
No installs, no setup — sign in, send it something, and it starts building.
This is the renewal my landlord sent over.
Read it. Twelve-month renewal, rent up 4 percent, and you have to sign by September 12. I've filed it with your housing documents — want a reminder before the deadline?
A week before.
Done — reminder set for September 5.
— two days later, in a different app —
When do I have to sign the lease?
September 12 — from the renewal you sent me on Telegram Tuesday. Your reminder is already set for the 5th.
It read a scanned lease, filed it, set the reminder — and still knew everything two days later, in a different app, unprompted. That conversation runs identically on a machine in your own house.
Most of what a personal assistant should do, you already pay for in pieces — spread across companies whose interests are not yours, holding data you cannot move. iSneps runs those functions itself, over one memory, and can run all of them on hardware you own.
Being straight about the edges: it does not host your mail server, it does not replace your bank, and moving a decade of photos onto your own box is a weekend, not a click. What it does mean is that the index, the memory and the answers are yours, and they stay yours if you leave.
The industry shipped "memory" as a sticky note of preferences. That is not memory, it is personalization. iSneps ingests the actual record of your life and answers across all of it at once — which is only tolerable because that record belongs to you.
And you just ask, in plain words:
Find photos of Sarah
Summarize the PDF I uploaded yesterday
How have I been sleeping this week?
A memory that only recalls is a filing cabinet. This one holds the pen.
It runs the errand inside the conversation. Sets reminders and calendar events, reads and sends email, controls your smart home, searches the web, generates images, and can speak its replies aloud.
Turn off the living room lights
It hosts your research. Ask a hard question and it dispatches a council of experts that comes back with a cited report — saved to your library, on your infrastructure, becoming part of what it knows about you. Nobody meters it and nobody reads it.
Compare the top three local LLM inference engines
It edits you a newspaper. A live personal front page, built fresh every time you open it and shaped by what you actually read — not by what a platform wants you to see.
What's in my news this morning?
240+ built-in abilities · 11 model providers with automatic failover · 20 Telegram Mini Apps · 13 Obsidian vault tools
Every other assistant is a front end for exactly one company's model. When that company changes its pricing, its policies or its mind, you have no move. Here the memory is permanent and the intelligence is swappable — they are separate parts on purpose.
Gemini, Claude, OpenAI, Grok, Venice, DeepSeek, Qwen, Moonshot, MiniMax, Zhipu and local models through Ollama. If one is down, billing-blocked or refuses, the next one answers — automatically, mid-conversation.
Point it at a model running on your own GPU and the platform works the same way: same memory, same tools, same surfaces. Slower, entirely offline, and nothing leaves your network.
The claim is only worth something if you can check it, so the parts that make self-hosting real are published rather than described.
Your own assistant in Telegram and Discord. One executable opens a setup wizard: pick a provider, paste your keys, name your bot, and it is live — with memory, voice notes, images, web search, reminders and Home Assistant control. Everything runs on your machine.
The routing layer, extracted and released on its own: model catalog, task-typed routing, fallback chains, a circuit breaker, and cost and usage tracking. The piece that makes eleven providers behave like one. Useful in your own projects, whatever you are building.
Desktop
The full platform on your own machine rather than a browser tab. It runs; there is no public installer yet.
More of the platform gets published as each piece is ready to stand on its own. The hosted service and the self-hosted kit are the same software — the hosted one just means you do not have to be your own sysadmin on a Tuesday.
There is no iSneps world to move into. Tell it something in one place, ask in another, and it has already caught up. Statuses are literal — Live is usable today, Beta runs with rough edges, Experimental works but is not under active development, In development does not exist yet.
Your conversations, files and memory live in per-user databases scoped to your account — not pooled, not shared, not mined. There is no advertising business here to feed.
Because the self-hosted platform is the same software, your data has somewhere to go. Leaving is a migration, not a negotiation — which is the only version of "your data is yours" that means anything.
For developers: an OpenAI-compatible endpoint (/v1/chat/completions) lets any OpenAI SDK point at your iSneps.
Stated plainly, because a plan you can hold someone to is worth more than a roadmap. None of the four below exists yet; when one ships it moves up the page and gets a status badge like everything else.
The goal is not a suite, a marketplace or a subscription tier. It is one personal AI platform that is genuinely good, that you can hold in your hands, and that does not need anyone's permission to keep working.
Send it something today. Ask about it anywhere, any day after. It will already know — and if you would rather it all ran in your own house, that version is on GitHub.