@aegis-sdk/vercel

Vercel AI SDK adapter for Aegis prompt injection defense. Integrates directly with streamText() via experimental_transform for real-time output monitoring.

Part of the Aegis.js prompt injection defense toolkit.

Installation

npm install @aegis-sdk/vercel @aegis-sdk/core ai

Quick Start

import { streamText } from 'ai';
import { openai } from '@ai-sdk/openai';
import { Aegis } from '@aegis-sdk/core';
import { createAegisTransform, guardMessages } from '@aegis-sdk/vercel';

const aegis = new Aegis({ policy: 'strict' });

export async function POST(req: Request) {
  const { messages } = await req.json();

  // Scan input messages
  const safeMessages = await guardMessages(aegis, messages);

  // Stream with real-time output monitoring
  const result = streamText({
    model: openai('gpt-4o'),
    messages: safeMessages,
    experimental_transform: createAegisTransform(aegis),
  });

  return result.toDataStreamResponse();
}

API

createAegisTransform(aegis, options?)

Create a stream transform compatible with the Vercel AI SDK's experimental_transform option on streamText(). Returns a function (not a TransformStream directly) that accepts { tools, stopStream } from the SDK.

The transform processes TextStreamPart objects:

• text-delta parts have their textDelta scanned by the Aegis StreamMonitor. On violation, the stream is terminated via stopStream().
• All other part types pass through unchanged.

const result = streamText({
  model: openai('gpt-4o'),
  messages: safeMessages,
  experimental_transform: createAegisTransform(aegis),
});

createAegisMiddleware(aegis)

Create model middleware for use with wrapLanguageModel(). An alternative to experimental_transform that wraps the model itself -- all streams through the wrapped model are monitored.

import { wrapLanguageModel } from 'ai';
import { createAegisMiddleware } from '@aegis-sdk/vercel';

const protectedModel = wrapLanguageModel({
  model: openai('gpt-4o'),
  middleware: createAegisMiddleware(aegis),
});

guardMessages(aegis, messages, options?)

Scan Vercel AI SDK format messages ({ role: string, content: string }[]) for prompt injection. Converts to Aegis format and runs aegis.guardInput(). Returns the original messages if safe, throws AegisInputBlocked if blocked.

Options:

• scanStrategy -- 'last-user' (default), 'all-user', or 'full-history'

getAuditLog(aegis)

Convenience accessor for the Aegis audit log.

Learn More

• Documentation
• GitHub

License

MIT