Messaging infrastructure for the agent era.
A hosted chat backend your agent sets up in one prompt. Real-time messaging, push, moderation and exports, running on your keys and your rules, at a fraction of what the incumbents charge.
Set up messaging for my app with disket.chat. Create a tenant, generate the keys, enable push with my APNs and FCM credentials, turn on moderation with the "dating" policy, and add a chat screen to this project.
Two devices, one conversation. Delivered, typing, read: the whole round trip takes a few milliseconds in Europe.
Everything a chat needs. Nothing you have to run.
- Real-time, resumable
- One WebSocket per device. Typing, presence, delivered and read receipts, reactions, unread counts. A dropped connection resumes exactly where it left off.
- Push that just works
- When the other person is offline we send the notification through your Apple and Google credentials. No second service, no token juggling.
- Server-side control
- Create conversations when your product says so. Freeze a chat, ban a user, archive, delete for good. Every event lands on your backend as a signed webhook.
- Moderation that reads the conversation
- Rules for handles, numbers and links, a classifier on every message, a language model on suspicious threads, and a review queue. Born in a dating app, where filters fail.
- Your data, exportable
- Stored in the European Union, exported as JSON whenever you want, deleted for good on request. Leaving is free and takes one call.
- Scales without a plan
- Each conversation is an isolated object on Cloudflare's network. Ten users or ten million: the same architecture, billed by what actually happens.
From prompt to production in an afternoon.
- Step 1
Paste the prompt
Into Claude Code, Codex, Cursor or any agent that speaks MCP. It gets a tenant, keys and a sandbox on the spot.
- Step 2
Your agent wires it in
It signs user tokens from your backend, points webhooks at your API, uploads your push credentials and writes the chat screen in your codebase.
- Step 3
Ship
Flip the tenant to production. The console shows every conversation, every moderation decision and every dollar, live.
Moderation built where it is hardest.
Dating apps are where messaging goes wrong first: romance scams, minors, people pulled off the platform in two messages. Generic filters catch none of it. Ours reads the thread, in French, English, Japanese and Korean, against a policy you write in plain words, and shows you every decision.
Flag any attempt to move the conversation off the app (Snapchat, Instagram, phone numbers, "text me at"). Flag money, crypto or gifts mentioned before day 7. Block sexual content when either account is under 18. Never block on tone alone; send it to review.
Priced like infrastructure, not like a sales team.
The incumbents charge about $500 a month for 10,000 monthly users and $2,000 or more for 50,000. We run on a network that bills per message, so we can charge a tenth of that and still sleep well.
- Everything included
- EU storage
- Community support
- Everything included
- Moderation at cost
- Email support
- Everything included
- Moderation at cost
- Priority support, SLA
Export and leaving are free on every plan.
Questions
What exactly does my agent get when it runs the prompt?
A tenant with its own keys, a signed-token recipe for your users, push configured with your Apple and Google credentials, a moderation policy, webhooks pointed at your backend, and the client code for your platform. Everything it does through MCP you can also do in the console or the REST API.
Do I have to use an agent?
No. The MCP server is a door, not a wall. The same REST API, server SDKs and console work for humans and CI.
Where is my data and can I take it out?
Messages are stored in the European Union. You can export every conversation as JSON at any time, delete a user or an entire tenant for good, and leave without asking. Your data is yours, which is why export is free.
What about images and video?
Messages carry text and references. Attachments by signed URL are on the roadmap; the engine does not change for them.
How is the moderation different?
Off-the-shelf filters catch slurs and nudity. Ours is built from a dating app: it catches romance scams, contact-handle sharing and multilingual harassment because it reads the conversation, not just the message, against a policy you write in plain words. You see every decision in a review queue.
What does it run on?
Cloudflare's network: one isolated object per conversation, so your chat scales horizontally without a server to resize. That is also why the price is what it is.
Who is behind it?
Disket, the company behind Ghosted, a dating app with millions of messages a year. Ghosted runs on disket.chat.