@caesura-io/ai-sdk

Asynchronous, non-blocking recommendation injection for the Vercel AI SDK.

Caesura listens to your agent's dialogue and pushes short, real-time recommendations ("analysis") into the model's context before the next call — without blocking the conversation. It plugs in as a standard AI SDK language model middleware.

Status: early development. API is not yet stable.

Install

npm i @caesura-io/ai-sdk ai

Quick start

Wrap your existing model with caesuraMiddleware — the only changes to your code are the highlighted lines:

 import { wrapLanguageModel, generateText } from 'ai';
 import { anthropic } from '@ai-sdk/anthropic';
+import { caesuraMiddleware } from '@caesura-io/ai-sdk';

-const model = anthropic('claude-sonnet-4-6');
+const model = wrapLanguageModel({
+  model: anthropic('claude-sonnet-4-6'),
+  middleware: caesuraMiddleware({
+    baseUrl: 'https://dev.caesura.io',
+    // apiKey auto-read from CAESURA_API_KEY if omitted
+  }),
+});

 const result = await generateText({
   model,
   messages: conversation,
+  providerOptions: { caesura: { conversationId: sessionId } },
 });

Credit Usage Reporting

You can request credit-usage metadata on every analyze call and receive the reported value via the onCreditUsage callback.

+import { caesuraMiddleware, createCreditMeter } from '@caesura-io/ai-sdk';
+
+const meter = createCreditMeter();

 const model = wrapLanguageModel({
   model,
   middleware: caesuraMiddleware({
     baseUrl: 'https://dev.caesura.io',
+    onCreditUsage: meter.record,
   }),
 });

+// Query credit metrics later
+console.log('total credits consumed:', meter.total());
+console.log('credits by conversation:', meter.breakdown());
+console.log('retained credit events:', meter.events());

[!NOTE] In async mode, the onCreditUsage callback fires out-of-band as soon as the asynchronous analyze call completes, decoupled from the synchronous generateText response.