oll.am is a pay-once AI workspace that runs any model, open-weight or frontier, with on-device privacy. No monthly tax. Choose how far your words travel: on your device, on private servers, or out to the world's best AI.
ChatGPT charges $20/month for one model from one company. Claude charges $20/month for one model from one company. So people stack four tools, pay $66/month, and many end up cancelling and re-subscribing just to keep the bill in check.
Local AI already solved the cost and privacy problem, but Ollama and Open WebUI built it for developers, behind Docker. oll.am closes the gap: any model, a beautiful workspace anyone can use, privacy you control, and you pay once.
Every other AI tool decides for you. oll.am makes it your call, one tap per conversation.
Runs entirely on your machine, in-browser via WebGPU or in the desktop app. Nothing leaves. The thing no mainstream AI offers.
Open-weight models on private, hosted infrastructure. Your words aren't used to train any model. Private, auditable, fast.
Claude, GPT and the best of the rest, when you want maximum quality. A deliberate, opt-in trade. Your data leaves your device, and you chose it.
Defaults to the safe tier. Switching is one tap, and Turbo always asks first. oll.am never quietly sends your words somewhere you didn't choose. Under the hood, each service owns its own private database — data boundaries are enforced by the architecture, not convention.
Say the task; oll.am routes it to the right model at the right privacy level, and shows you the cost before it runs.
"Rewrite this paragraph to sound more academic" → Groq Llama 4 · fast, sufficient · $0.001 "Generate an architecture spec from this braindump" → Claude Sonnet · complex reasoning · $0.05 "Generate 8 professional photos in different styles" → Replicate Flux LoRA · image gen · ~$2.00 "Summarize this confidential document" → Local Llama 8B (WebGPU) · 🔒 Vault · $0.00
Anything private routes to Vault automatically, and oll.am suggests it rather than forcing it ("this looks like health data, keep it on-device?"). Complex, non-sensitive work goes to the cloud model that does it best. You set your preference once; oll.am handles the rest.
| ChatGPT | Claude | Ollama | LM Studio | oll.am | |
|---|---|---|---|---|---|
| Web UI, no setup | ✓ | ✓ | ✗ | ✗ | ✓ |
| Any model | ✗ | ✗ | ✓ | ✓ | ✓ |
| On-device / private | ✗ | ✗ | ✓ | ✓ | ✓ |
| Hosted open-weight tier | ✗ | ✗ | ✗ | ✗ | ✓ |
| Image gen + your face | DALL·E | ✗ | ✗ | ✗ | Flux LoRA |
| Code sandbox | ✗ | Artifacts | ✗ | ✗ | ✓ |
| Built for non-technical | ✓ | ✓ | ✗ | ~ | ✓ |
| Cost | $20/mo | $20/mo | Free·CLI | Free·app | Pay once |
AI produces things: images, documents, code, drafts. You see them front and center. Not another chat box. A creative workspace.
Five modules, one workspace. Each turns a prompt into something you can download, send, or ship.
Train on your own face, then generate professional photos, product shots, and portraits in eight styles, download-ready.
In ProChat with any model, then rewrite, clarify, and translate, your voice kept. An honest writing tool, not a detector-beater.
In StandardGenerate specs, CVs, and cover letters with a quality score, so you know what's strong before you send it.
In StudioA coding agent in a sandbox: a fresh container per session, with a live preview of what it builds.
In StudioBackground tasks that run while you don't: discover threads, draft replies, keep work moving on its own.
In StudioLock in founding pricing, about half of launch price, before it ships. No subscription, ever.
Pro bundles what would cost ~$35/month in subscriptions (image gen, rewriting, a photo session) into a single payment you keep. It pays for itself in about two months, then it's yours.
Not ready to commit? Join the waitlist for CHF 1 → A founding spot held, early access the day it ships.
oll.am isn't a demo. It's a live, multi-service platform, and the design is public. Two documents cover how it's built and how it ships — read either to look under the hood.
The full system: a frozen auth·billing·email core, a provider-routing model gateway, and thin product services over HTTP — with the running-services and deployment diagrams.
Architecture + diagrams →The delivery pipeline: contract-tested CI, a preflight gate, and dependency-ordered, health-gated deploys — decoupled from version control, so a merge never deploys.
CI/CD design →I'm Samuel Alemu, a software engineer in Zürich, and I build oll.am end to end — the product and the platform under it. Every decision, audit, and spec goes public the day I make it, so you can read the whole story before you spend a cent.
This is a founding pre-order. oll.am ships in Q3 2026. If it doesn't, you get a full refund. No fine print.