Chat API · SDK · llms.txt · Free
Tell your AI. Get high-performance chat.
Your part ends at signing up and getting one API key. Your AI does the rest, from creating the app to wiring up chat. Example prompts are ready to paste.
$ npm install @kispi/chat
import { createChatClient } from '@kispi/chat'
const chat = createChatClient({ key: 'pk_3f9a1c7e5b', token })
await chat.connect()
const room = chat.roomByKey('lobby')
await room.subscribe()
room.on('messages', render)
await room.send({ text: 'hi everyone 👋' })
// 00 — LIVE
Try it right here
This is the real engine, not a mock. No sign-up — say something, then flip between Slack-style threads and KakaoTalk-style replies. Same room, same data; only the UI changes.
Live demo · anyone can jump in and talk
Same room, same data — just switch the UI
Connecting…
// 01 — IN PRODUCTION
Already running in production
The chat in these services runs on this engine.
// 02 — USE CASES
Any kind of chat you need
Open rooms anyone can join, private rooms by invite, one-on-one DMs, and channels only admins post to. Mix them however your product needs.
Community open chat
Anyone can drop in and talk, and you can show how many people are in the room.
- Public rooms
- Who's online
Support & direct messages
Customer support, marketplaces, dating matches — private talk between two people, with read receipts.
- DMs
- Read receipts
Team messenger
Slack-style invite-only rooms. Side conversations move into threads.
- Private rooms
- Threads
Announcement channels
Like a Telegram channel: admins post, everyone reads. With webhooks on, new posts reach your server, so app notifications pick up from there.
- Channels
- Notification hook
Live stream chat
A crowd reacting to the same moment. You can block abuse before it reaches the room.
- Reactions
- Pre-publish check
In-game & in-app chat
Guild chat, party chat, match lobbies. It plugs into the login you already have.
- Private rooms
- Mentions
// 03 — FEATURES
Everything a chat needs, already here
-
Read receipts & unread counts
Who read what, and an unread count on every room.
-
Typing indicator
See who is typing in a room, live.
-
Emoji reactions
Tap to react, and see who reacted with what.
-
Replies & threads
Answer a message in place, or branch off into a thread.
-
Photos & files
Attach photos and files to messages. They stay in the storage you already use.
-
Mentions
@ someone to get their attention.
-
Who's online
See who is in the room right now.
-
Edit & delete
Fix a typo, or take a message back.
-
Filter before sending
Turn on the before_publish webhook and each message asks your server before it reaches the room. Allow, rewrite or block it — your rules decide.
-
Notification hook
With webhooks on, every new message is reported to your server. Send the push from there.
// 04 — SEQUENCE
Easy to add, built to hold up
Every message gets a number that is unique in its room, and every subscriber receives them in the same order.
Drop the connection, lose nothing
A tunnel is fine. The moment the connection is back, the SDK fills in whatever arrived while you were gone.
Before the drop
Back online — a gap is spotted
Filled in automatically
Order never breaks
Hundreds of people can type at once and everyone still sees the same order. Nobody gets their own version.
We run the servers
The engine, the database and the deploys are on our side. Your code looks the same as it did with your first ten users.
A small SDK
TypeScript with zero runtime dependencies. Same code in the browser and on Node.
The structure, the numbers we measured and what we did not measure are in the architecture and benchmarks doc.
// 05 — FIT
Where it fits, and where it does not
Check the fit before you build. If something on the "does not fit" side is central to your product, pick something else.
A good fit
- Real-time chat next to a web community or web content.
- Side projects, MVPs and early-stage startups.
- Web-first one-on-one support and marketplace DMs. Push goes out by webhook, then on to FCM.
- Teams that were about to build their own Socket.IO server.
Not a fit
- Products that are native mobile apps first. There is no native SDK and no built-in push.
- Products where chat is the core of revenue and you need SLA compensation or compliance. The uptime figure is an operating target with no compensation.
- Places that need audits or certification, such as healthcare and finance.
- Teams that want a finished chat UI kit right away. The SDK does not give you screens; you draw them yourself.
// 06 — PLAN
Free, and you can leave any time
$0
Basic is free. No card and no trial period.
No concurrency or MAU cap set by pricing
Each app has safety limits (for example 10 concurrent connections per user), but pricing does not cap usage.
Export any time
One console button downloads all users, rooms and messages as an NDJSON file. The data is yours.
Pricing and per-app limits SLA: uptime target, backups and notice periods
// 07 — START
Three steps and you are live
- 01
Sign up, get an account key
Sign up with an email code, create one account key (ck_) in the console and put it in .env. That is the only human part.
- 02
Hand your AI a prompt
Copy a prompt from the docs and paste it. Your AI reads llms.txt, creates the app through the console API and fills .env with the publishable key (pk_) and secret key (sk_).
- 03
Chat is wired up
Your AI adds ticket issuing to your server and the SDK connection to your UI. Your login stays as it is; without one, start with guests.
$ npm install @kispi/chat
One API key. Your AI does the rest.
Your part is signing up and getting a key. Your AI handles app setup and chat wiring — with example prompts ready to go.