Chat, work
& code, in private.
Anonymous by default · messages not stored on our servers · open-weight models · what is and is not protected
A complete assistant, for your everyday.
Understand, draft, compare, decide. The best open models, served by European GPUs.
An agent that delivers the file, not just the answer.
Up to 8 chained steps: research, calculation, file generation. Kowork runs Python, produces your Excel, Word, PDF and slide decks, and connects to your tools (Linear, GitHub, Notion, Slack).
An agent that writes, runs, and verifies.
Kode reads your project, proposes a fix, runs the tests, and proves it works.
What happens to a question,
in four steps.
No account needed
Your question travels over TLS with no account or identifier attached. The computer that answers reads it, as any online service would.
A peer GPU answers
A computer lent by its owner hosts the model, never a Korai datacenter.
Anonymous tokens pay
One anonymous token covers the request. Neither your card nor your email travels with it.
Tokens are not linked to you
Korai cannot match a spent token to the one it signed. Your IP address and a random browser id can still link chats from one browser to each other.
For work, for life,
and everything else that stays private.
Writing
From rough notes to finished text: letters, contracts, articles, messages. Without your drafts training someone else's model.
Fiduciary & legal
Case analysis on open-weight models. Check your professional-secrecy obligations first: the computer that runs the model reads what you send.
Code
Write, debug, understand a codebase. The Kode agent runs the tests and proves it works.
Documents
Summarize, extract, cross-reference your PDFs and spreadsheets. Files are read in your browser and travel with your message, never stored on Korai's servers.
Health
Ask the questions you don't want in any advertising history. Korai builds no ad profile and does not store your messages.
Research
Explore a topic in depth with deep mode: the strongest reasoning model works through it step by step.
What Korai guarantees,
and what it doesn't.
Three things we can state plainly today. The limits are listed on the About page.
Encrypted in transit
TLS on every hop, from your browser to the host computer. Not end to end: the servers on the way read the message while it passes.
Tokens, not accounts
In anonymous mode you pay with blind tokens and no account is sent with your message. Chats from one browser still share a random id.
No stored messages
Korai's servers do not store the text of your messages. Your history stays in your browser unless you turn on encrypted sync.
Always the best
open-weight models on the market.
Three profiles cover speed and everyday reasoning (Qwen 3.8), multimodal (Gemma 4), and the deepest reasoning in Ultra Deep (Kimi K3). The fleet is audited every 30 days against public benchmarks and updates itself automatically across the whole network.
Qwen 3.8-27B
Alibaba · Apache 2.0
GPQA Diamond 89.2 · LiveCodeBench v6 90.3
Gemma 4 26B A4B
Google DeepMind · Apache 2.0
GPQA Diamond 82.3 · LiveCodeBench v6 77.1
Kimi K3
Moonshot AI · 1,000,000 token context
Artificial Analysis Intelligence Index 57 · #1 Frontend Code Arena
Your GPU has value.
The reef proves it.
Korai owns no datacenter. The network is made of machines like yours: a Mac, a tower with an NVIDIA card. You host open models, you get paid per second of compute.
The app detects your hardware, loads the right models, and the network does the rest. Requests arrive without the requester's account or IP address, but you can read the text your machine processes.
Simple pricing,
built for Europe.
Four tiers, from a free no-signup entry to intensive professional use. These are the tiers in effect, not a projection.
To try Korai with no commitment.
- ~3.6M tokens/month across every Korai model
- No signup needed to start chatting
- Anonymous by default, no exceptions
- Korai Kode command-line tool
- Community support
For daily use.
- 30M tokens/month across every Korai model
- Access to deep mode (extended reasoning)
- Agent mode (Kowork) and MCP connectors
- Korai Kode command-line tool
- Email support
For intensive professional use.
- 100M tokens/month across every Korai model
- Korai Reef (Ultra Deep) capped at 30 requests/day
- Korai Kode with persistent project memory
- Agent mode (Kowork) and MCP connectors
- Priority email support
For the most demanding use cases.
- 310M tokens/month across every Korai model
- Korai Reef (Ultra Deep) with a generous quota
- Korai Kode with persistent project memory
- Early access to new models and features
- Priority email support
Your AI governance, taken care of. Your teams use frontier LLMs with no risk of data leakage.
- Turnkey AI governance: shadow IT becomes managed use
- No message text stored on Korai's servers
- Dedicated deployment (private cluster or on-prem), data hosted in France only
- SSO integration (SAML, OIDC), GDPR / NIS2 compliance audit
- Contract, SLA and dedicated support
The same rate, per million tokens.
Developer integrations are billed per million tokens, no tier, no commitment.
| Model | Input (per M) | Output (per M) |
|---|---|---|
| Korai CurrentQwen 3.8 27B | €0.10 | €0.50 |
| Korai GemmaGemma 4 26B A4B | €0.20 | €0.50 |
| Korai ReefKimi K3 · sovereign EU datacenter | €1.00 | €4.00 |
Korai names are stable: the backend is always the best open-weight model available, the catalog rotates.
Korai Current: up to 27% cheaper than equivalent market API pricing on input, with EU sovereignty and privacy on top.