Open source · Electron browser

Every page you visit
has a room.
You're never the only one in it.

Paperkite is a browser with a chat panel built into every tab. Anyone else on the same URL right now can talk to you there. No accounts, no invites, no separate app. Bring an AI agent into the conversation, or run the server yourself.

Download Paperkite View source

MIT licensedmacOS · Windows · LinuxRead the wire protocol →

the four-day week, one year in

🔒fieldnotes.example/four-day-week-one-year-in

4 here now

Page chat

🇯🇵yukianyone here from a team that tried this

🇧🇷marianawe switched in january, mixed results so far

🇩🇪lenathe numbers in section 3 match what I have seen

🇳🇬yousame page, three different countries

someone is typing

Open any page

No setup. Paperkite hashes the URL into a room the instant a tab loads it.

Land in its room

Everyone else currently on that same URL, anywhere, on any connection, is already there.

Talk, or loop in an agent

Chat with the room, or hand the page to Claude, GPT, Gemini, or a local Ollama model.

What's in the box

What Paperkite ships with.

Bring your own MCP servers

Add any Model Context Protocol server to the browser. Your agents can call its tools while browsing: file systems, search, internal APIs, whatever you point them at.

Ships with its own MCP server

Paperkite exposes the page you're on, its content, the room, and the chat, as MCP tools of its own. External agents can act on what you're browsing instead of only discussing it.

Live audio translation

Whisper-powered live translation for audio on the page: calls, videos, streams. Swap in your own custom models if you want.

Whole-page translation

Translate the page you're reading with open-source or commercial models, whichever you prefer. No lock-in to one vendor's quality or pricing.

Fully open source

Browser, chat protocol, and server are all MIT licensed. Read the wire format, run your own fork, audit every byte that leaves the app.

No walled gardenNo forced accountsNo telemetry by defaultSelf-hostable end to end

AI agents, in the room

Chat with an agent about the page you're on.

Open the agent panel next to any tab and ask about what's in front of you. Paperkite ships adapters for the major hosted providers and for open-source models through Ollama. Pick a provider per conversation, keep your keys local, and switch anytime.

ClaudeAnthropicGPTOpenAIGeminiGoogleOllamaany local model

Agent · Claude

summarize the top comment thread on this page

Most replies point to the same finding: teams cut meeting time in half but individual output held steady. One thread links the internal report the study is based on.

chat-service

$git clone

go build ./cmd/server

CHAT_JWT_SECRET=... ./server

#HTTP :8080 · gRPC :50051

Run your own

Don't trust our server. Run yours.

The chat backend is a small, stateless Go service. Deploy it in a container, on a VPS, or on your laptop for a LAN. Point Paperkite at it from Settings, and every room on that instance runs on your infrastructure.

Read the connection details

PROTOCOL.md