> This page is for Server SDKs. > Fetch clean Markdown by appending `.md` to any page URL under https://signalwire.com/docs or requesting it with the HTTP header `Accept: text/markdown`. The root index at https://signalwire.com/docs/llms.txt lists the available documentation indexes. # prepare > Validate a browser request and build the call to the chat service. [chatgateway]: /docs/server-sdks/reference/typescript/agents/chat-gateway [rawpost]: /docs/server-sdks/reference/typescript/agents/ai-chat-client/raw-post [router]: /docs/server-sdks/reference/typescript/agents/chat-gateway/router [readhandle]: /docs/server-sdks/reference/typescript/agents/chat-gateway/read-handle [visiblemessages]: /docs/server-sdks/reference/typescript/agents/chat-gateway/visible-messages [lastactivity]: /docs/server-sdks/reference/typescript/agents/chat-gateway/last-activity [createconversation]: /docs/server-sdks/reference/typescript/agents/ai-chat-client/create-conversation [log]: /docs/server-sdks/reference/typescript/agents/ai-chat-client/log [end]: /docs/server-sdks/reference/typescript/agents/ai-chat-client/end Validates a browser request and builds the call to forward. [`router()`][router] uses it; call it yourself to serve the gateway from a framework other than Hono. It checks the key, then the origin. Everything the browser could use to widen its own access is refused or replaced here: `method` must be one of `start`, `chat`, `log`, or `end` (default `chat`), the conversation comes from a signed handle, and `config_url` is the gateway's. The one field forwarded is `user_meta_data`, after the checks described under [`ChatGateway`][chatgateway]. A chat message over 8 KiB of UTF-8 is refused with `413` before a conversation is created or a turn is counted. `prepare()` doesn't check the size of the request body. [`router()`][router] limits it to 64 KiB before parsing; apply a limit of your own when you call `prepare()` directly. A `log` or `end` request needs a handle. A `start` or `chat` without one creates a conversation, which counts against `maxNewConversations`; a `chat` counts against `maxTurns`. ## Parameters **`body`** `Record` — required The browser's request body: `{ method?, handle?, message?, user_meta_data? }`. --- **`opts`** `{ origin?: string | null; key?: string | null }` — required The request's credentials. --- **`opts.origin`** `string | null` The `Origin` header. --- **`opts.key`** `string | null` The presented publishable key. --- ## Returns `[string, Record, string | null]` -- a tuple of `[method, params, mintedHandle]`. `method` is the service method to call: `create_conversation` for `start`, `chat` for `chat`, `chat_log` for `log`, and `end_conversation` for `end`. `params` are its parameters. `mintedHandle` is the new handle when this call created the conversation, and `null` otherwise; return it to the browser. Send each method the way [`router()`][router] does: | `method` | Send it with | Answer the browser with | | --------------------- | -------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `chat` | [`rawPost()`][rawpost] | The response body, streamed through as it arrives | | `create_conversation` | [`createConversation()`][createconversation] | `{ greeting, status, timeout }` | | `chat_log` | [`log()`][log] | `{ messages, timeout, last_activity }`, with `messages` filtered by [`visibleMessages()`][visiblemessages] and `last_activity` from [`lastActivity()`][lastactivity] | | `end_conversation` | [`end()`][end] | `{ status: 'ended' }` | Never relay a `chat_log` response as it comes back. The full log carries the system prompt and tool traffic, which the browser must not see. ## Throws `GatewayRejection` when the request is refused: `401` for a bad key, `403` for a disallowed origin or a bad handle, `400` for a malformed request, `413` for an oversized message or metadata, and `429` when a cap was reached. ## Examples ```typescript import { AIChatClient, ChatGateway, GatewayRejection } from '@signalwire/sdk'; const gateway = new ChatGateway({ configUrl: 'https://bayview-taxi.example.com/dispatch', key: '', allowedOrigins: ['https://bayviewtaxi.example.com'], }); const client = new AIChatClient({ space: 'your-space' }); async function forward( method: string, params: Record, headers: Headers, ): Promise { const id = params.id as string; const timeout = gateway.effectiveTimeout; if (method === 'end_conversation') { await client.end(id); return Response.json({ status: 'ended' }, { headers }); } if (method === 'create_conversation') { const info = await client.createConversation(id, { configUrl: params.config_url as string, timeout: params.conversation_timeout as number | undefined, userMetadata: params.user_meta_data as Record | undefined, }); return Response.json({ greeting: info.initialMessage, status: info.status, timeout }, { headers }); } if (method === 'chat_log') { // Filter the log: the full chat_log isn't for the browser. const log = await client.log(id); return Response.json( { messages: ChatGateway.visibleMessages(log.messages), timeout, last_activity: ChatGateway.lastActivity(log.messages), }, { headers }, ); } // chat: stream the reply through as it arrives. const upstream = await client.rawPost(method, params); headers.set('Content-Type', 'application/json'); return new Response(upstream.body, { status: 200, headers }); } export async function relay( body: Record, origin: string | null, key: string | null, ): Promise { try { const [method, params, minted] = gateway.prepare(body, { origin, key }); const headers = new Headers(); if (minted) headers.set('X-Chat-Handle', minted); return await forward(method, params, headers); } catch (err) { if (err instanceof GatewayRejection) { return Response.json({ error: err.reason }, { status: err.status }); } throw err; } } ```