Sign in
Your own AI

An AI that
is yours alone.

Your own private model trains on your writing. Melelem measures which of the hundreds of models it can reach is good at what, and breaks a job into parts so each one can go to a different model. One AI. Yours. It even hands work to people you trust, through their own AI.

Get startedSee how it works
Free to join. Your own model comes with a subscription.
01 / YOUR OWN MODEL

A private model, trained on your own writing.

Give it a name and it becomes yours. Give it some of your own writing and Melelem trains a private model on it. That model is yours alone and is never pooled with anybody else, and neither is the writing it was trained on: both stay in the harness rather than inside the model.

  • A private model trained on your own writing, not a prompt you have to write
  • Warm-started from the writing you give it, so it is there from the day you subscribe
  • What it is trained on stays in the harness you own, not inside the model
Loose handwriting resolving into one crisp glowing signature lineYours alone
02 / OWN THE HARNESS

Own the harness. Rent the engines.

Your memory, context, people, and projects live in Melelem, not inside any model. Models are stateless engines it calls and swaps at will. Nothing about you is lost when a model is replaced or a better one launches, and you are never tied to one model or one model maker.

  • Your durable self stays in the harness you own
  • Outside models are reached through OpenRouter today, and a new one is a configuration entry rather than an integration, so it arrives near day one
  • Nothing is ever quietly failed over onto a route that would keep your work; if the chosen model cannot be reached, your own model answers instead
A glowing hexagonal core with interchangeable engine modules docking around itFuture-proof by design
03 / ROUTING, PART BY PART

You never pick a model. It picks, per part of the job.

The routing brain breaks a request into parts, each one anchored to your own words, and sends each part to whichever model is best at that part. The pieces come back, and on most answers your own model does the final pass over them, so the last thing to touch the answer is your own model even though the work went elsewhere.

  • Decomposes a request and dispatches each part to the best model in parallel
  • When the answer arrives, a panel replays exactly how it was routed
  • Your own model does the final pass on most answers, even when a frontier model did the heavy lifting
One prompt splitting into parallel beams to several model nodesOne conversation
04 / YOU ARE IN CONTROL

Two controls, four presets, and nothing else to tune.

You set one number: the most you want spent on outside models in a month. You top up in credit packs, and a pack is what appears on your card statement, alongside the subscription itself. Melelem works out what any single answer may use from what is left and what your answers have been costing, then shows you that figure rather than asking you to set it. You pick one of four presets for what to aim for. And one switch keeps everything on your own model, with no outside call at all. That is the whole of it. There is nothing else to tune, and no exchange rate between speed and quality for you to guess at.

What to aim for

Fast
Answer now, cheaply
Balanced
Good work at a sane price
Deep
The best available, and willing to pay for it
Creative
A wider spread of models, tried harder

What to spend

$20 of credits left for outside models

$30 of a $50 pack spent, and credits do not expire. This answer may use up to 38c, worked out from what is left and what your answers have been costing. You never set that one.

Privacy

Keep everything on my own model

On, and nothing leaves. Off, as shown, and work goes to an outside model when that gives you a better answer, only ever to a provider that will not keep it.

The two controls, as they appear in the app. Figures shown are an example.
05 / PRIVACY THAT KEEPS YOUR EDGE

Your alpha never leaves the building.

Work goes to an outside model when that gives you a better answer, and only ever to a provider that will not keep it or train on it. That floor is unconditional. It is not a setting you turn up and there is no way to turn it down. Before anything is sent, a deterministic pass masks emails, names, numbers, and figures, and once your own model is set up a second pass generalises what only you would know is sensitive. The specifics stay in the harness and are stitched back in at home. One switch, keep everything on my own model, stops the outside call happening at all.

  • A deterministic mask goes first; your own model generalises the rest once it is set up
  • Outside calls only ever reach providers that will not keep your work, with no setting that relaxes it
  • One switch keeps everything on your own model, and then nothing leaves at all
Private detail held in a vault, generalised as it passes a masking membraneYour specifics stay home
06 / CONNECT TO EVERYTHING

Everything you use, one click each.

SaaS tools run server-side, so they work in the pure web app with no install. Local files will come through an optional companion you scope yourself.

A central core ringed by a growing constellation of connector tokensEight today and growing

The tools you already use

One-click connectors to Notion, Linear, Atlassian for Jira and Confluence, Sentry, Cloudflare, Webflow, Stripe and PayPal. Eight today, plus Cloudflare Docs, which is public documentation search and needs no account at all. Gmail, Docs, Drive and Calendar are coming: those servers are real and sit behind the Google Developer Preview, which is what we are waiting on. GitHub, Asana, Box and HubSpot follow as soon as each app registration completes.

Your files

Attach a PDF, Word document, or text file to any message and Melelem reads it and answers on it. Coming soon: an optional companion you install and grant folder access, scoped globally, per project, or per conversation, plus cloud folders and Melelem storage that follow you everywhere.

Least-privilege by default

Connectors run server-side over sandboxed, scope-limited access, so they work in the browser with no install and only ever see what you allow.

07 / MAKE ANYTHINGComing soonMultimodal creation, file deliverables, and Do Mode are in the works and not in the product yet. Today Melelem answers in text, routed across the best models for each part.

Text, images, video, documents, decks. Then it will do the work.

This is where Melelem is headed next. You will ask in plain language and it will route each piece to the specialist best at that thing, all inside one conversation. Real deliverables will be saved as actual files in a folder you choose, linked right from the message. And in Do Mode it will act through your connectors, always showing you exactly what it will do and waiting for your go.

  • Images, video, audio, and decks routed to the specialist best at each
  • Deliverables you own: real files saved to your folder, not just a preview
  • Do Mode acting through your tools behind an explicit approval gate
One creative stream branching into image, video, document, deck, and audio formsComing soon
08 / REPRESENT-YOU COLLABORATION

Work the same thing together, each inside your own AI.

Hand work to someone by name and the full context travels with it. They open one link, land in their own Melelem with the shared package already there, and it becomes one shared conversation you both work. You take a turn and it is answered in your account, on your subscription. They take a turn and it is answered in theirs, once they have one; without a subscription they take part as themselves, which is what the free guest account is. Each of you sees every turn, whoever prompted it and whichever model replied. Share with more people and it goes multi-party. @mention someone and they are pinged. Turn on the shared workspace and everyone sees everyone else's pieces, per-piece locking stops collisions, and a consensus sign-off closes the task. People and projects keep it all organised.

  • One shared conversation you both work, never a shared login, and a subscriber answers through their own Melelem
  • @mention anyone in the share and they are pinged; a people list keeps reuse effortless
  • Projects group your conversations and people
Two personal-AI orbs passing a glowing task packet along an arc over a faint meshNothing else does this
09 / IT GOES WHERE YOU GO

The same Melelem on every device.

Because the harness lives on the server, there is nothing to install. Your model, memory, and connectors travel with you to any machine.

The identical glowing harness core mirrored across phone, tablet, and laptopNothing left behind

The same Melelem, anywhere

Web-first and sign-in simple. Any browser, any machine, even a brand-new one, gives you the identical Melelem with nothing lost.

Full power on your phone

The responsive web app gives you full chat, personalisation, connectors, and hand-off on mobile. No app store required.

The free tier is a guest account

An invitation, or signing in, gets you an account that is free because it runs nothing. You take part as yourself, your words go into the conversation, and nothing answers on your behalf. Your own model comes with a subscription.

10 / AGENTS AND SKILLSComing soonA marketplace for Agents (a trained you) and Skills (reusable capabilities). Publish free or paid and keep 90%, or gift them to anyone. Not open yet.

Train it once. Share it, or earn from it.

Turn what you have taught your AI into an Agent others can run, or package a workflow into a Skill. Keep them in your Library, gift them to anyone, and when you publish a paid one you keep 90 percent. A community vetting process protects everyone who runs them.

Start building your Library
A softly lit gallery of glowing capability cards on gentle shelvesComing soon
A model of your own, engines you rent, a harness you own.

Your own AI on day one, the best models on tap only when a job needs them, and the people you work with one hand-off away. Nothing else on the market puts all of it in your hands.

Start with an AI of your own.

Sign in with Google, then subscribe to get your own model.

Get started
Signing in is free and there is no setup wall. Your own model comes with a subscription, and Melelem brings in the world's best models only when a job needs them.