Vercel Chat SDK adds XChat adapter with built-in end-to-end encryption for bots

Chat SDK gains an XChat adapter

Vercel has added an XChat adapter to its Chat SDK, allowing developers to build bots that take part in end-to-end encrypted one-to-one and group conversations on XChat. According to the Vercel blog, the adapter manages encryption, key management and signature verification automatically, removing the cryptographic plumbing that would otherwise fall on the bot developer.

Bots are not limited to replying, either. Vercel says a bot can message a user first, as long as that user has encrypted chat enabled and follows the bot. That opens the door to proactive or notification-style bots, but the capability stays gated on the user's own privacy setup rather than being unconditional.

How the integration works

The adapter follows the Chat SDK's existing pattern of plugging platform-specific transports into a shared Chat object. In the sample Vercel published, the bot imports Chat from the "chat" package, creates the adapter with createXchatAdapter from the "@chat-adapter/x/chat" package, and registers it under an "xchat" key. Message handlers such as onDirectMessage then receive a thread and an incoming message, and replies are posted back through the thread object — in the sample, the bot simply echoes the user's text with a "You said:" prefix.

The appeal of the abstraction is portability: bot logic stays the same while adapters translate it for each messaging surface.

Rendering fallbacks for a plain-text platform

XChat does not render markdown, so the adapter degrades rich output automatically instead of dumping raw syntax into the conversation. According to Vercel, URLs and @mentions are rendered as tappable links, tables are converted into ASCII code blocks, and cards fall back to plain text accompanied by a link preview.

Streaming is supported as well, and Vercel says it works through message edits, with content conveyed by updating an already-posted message.

Why it matters

End-to-end encryption is historically awkward territory for bots. When only the conversation endpoints hold the keys, any automated participant has to get key management and signature verification right, and mistakes in that layer are expensive. By folding that work into the adapter, Vercel makes it practical to run automated assistants inside private conversations without each team rolling its own crypto implementation — or, worse, weakening encryption so a bot can read the thread.

The integration also broadens the set of platforms where Chat SDK bots can run, and the automatic fallbacks mean developers who write markdown-heavy output for other surfaces do not need to maintain an XChat-specific branch of their formatting code. Some fidelity is lost — tables become ASCII blocks and cards lose their visual form — but the content still arrives in readable shape.

For teams already building on the Vercel ecosystem, the practical upshot is that an encrypted XChat bot is now closer to a configuration exercise than a standalone project. Vercel points developers to the adapter's documentation to get started.