39 lines
1.4 KiB
TypeScript
39 lines
1.4 KiB
TypeScript
|
import type { LanguageModelV3Middleware, LanguageModelV3Prompt } from '@ai-sdk/provider';
|
||
|
|
|
||
|
|
/**
|
||
|
|
* Creates middleware that adds Anthropic prompt caching directives.
|
||
|
|
* Marks the last system message and last user message with ephemeral cache control,
|
||
|
|
* reducing input token cost and latency on repeated calls.
|
||
|
|
*/
|
||
|
|
export function createAnthropicCachingMiddleware(): LanguageModelV3Middleware {
|
||
|
|
return {
|
||
|
|
specificationVersion: 'v3',
|
||
|
|
transformParams: async ({ params }) => {
|
||
|
|
const messages = [...params.prompt] as Array<Record<string, unknown>>;
|
||
|
|
|
||
|
|
// Find the last system message and last user message
|
||
|
|
let lastSystemIdx = -1;
|
||
|
|
let lastUserIdx = -1;
|
||
|
|
for (let i = 0; i < messages.length; i++) {
|
||
|
|
if (messages[i].role === 'system') lastSystemIdx = i;
|
||
|
|
if (messages[i].role === 'user') lastUserIdx = i;
|
||
|
|
}
|
||
|
|
|
||
|
|
const targets = [lastSystemIdx, lastUserIdx].filter(i => i >= 0);
|
||
|
|
for (const idx of targets) {
|
||
|
|
const msg = { ...messages[idx] };
|
||
|
|
msg.providerOptions = {
|
||
|
|
...(msg.providerOptions as Record<string, unknown> || {}),
|
||
|
|
anthropic: {
|
||
|
|
...((msg.providerOptions as Record<string, unknown>)?.anthropic as Record<string, unknown> || {}),
|
||
|
|
cacheControl: { type: 'ephemeral' },
|
||
|
|
},
|
||
|
|
};
|
||
|
|
messages[idx] = msg;
|
||
|
|
}
|
||
|
|
|
||
|
|
return { ...params, prompt: messages as unknown as LanguageModelV3Prompt };
|
||
|
|
},
|
||
|
|
};
|
||
|
|
}
|