Using prxy.monster with the Vercel AI SDK

The Vercel AI SDK (ai + @ai-sdk/* provider packages) can route supported OpenAI Chat Completions and Anthropic Messages calls through prxy.monster by setting the standard OPENAI_BASE_URL / ANTHROPIC_BASE_URL env vars.

Install

You probably already have these. If not:

npm install ai @ai-sdk/openai @ai-sdk/anthropic

Configure

Pick the provider you’re using.

export ANTHROPIC_BASE_URL=https://api.prxy.monster
export ANTHROPIC_API_KEY=prxy_live_xxxxxxxxxxxxxxxxxxxxxxxx

export OPENAI_BASE_URL=https://api.prxy.monster/v1
export OPENAI_API_KEY=prxy_live_xxxxxxxxxxxxxxxxxxxxxxxx

export ANTHROPIC_BASE_URL=https://api.prxy.monster
export ANTHROPIC_API_KEY=prxy_live_xxxxxxxxxxxxxxxxxxxxxxxx
export OPENAI_BASE_URL=https://api.prxy.monster/v1
export OPENAI_API_KEY=prxy_live_xxxxxxxxxxxxxxxxxxxxxxxx

Code change

None. Both @ai-sdk/openai and @ai-sdk/anthropic read those env vars automatically.

// Before AND after — no diff
import { anthropic } from '@ai-sdk/anthropic';
import { generateText } from 'ai';
 
const { text } = await generateText({
  model: anthropic('claude-sonnet-4-6'),
  prompt: 'Why is the sky blue?',
});

If you prefer to be explicit (or you’ve configured the provider via factory), pass baseURL:

import { createAnthropic } from '@ai-sdk/anthropic';
 
const anthropic = createAnthropic({
  baseURL: 'https://api.prxy.monster',
  apiKey: process.env.ANTHROPIC_API_KEY, // your prxy key
});

import { createOpenAI } from '@ai-sdk/openai';
 
const openai = createOpenAI({
  baseURL: 'https://api.prxy.monster/v1',
  apiKey: process.env.OPENAI_API_KEY,
});

Verify

curl https://api.prxy.monster/health
# → {"status":"ok"}

Or, if you’ve installed the CLI (@prxy/cli):

prxy doctor

What you get

• Infinite context (ipc) — your streamText / generateText calls compress old messages instead of dropping them.
• Semantic cache (semantic-cache) — repeated similar prompts return cached responses, near-zero latency.
• Pattern memory (patterns) — successful answers get learned and re-injected into similar future calls.
• Cost guards (cost-guard) — hard per-request / per-day budget caps.

Recommended pipeline for AI SDK apps

For a typical Next.js chat or RAG app:

PRXY_PIPE=mcp-optimizer,semantic-cache,patterns,ipc

For high-traffic production where cost is the priority:

PRXY_PIPE=exact-cache,semantic-cache,cost-guard,patterns

For agents that use a lot of tools (MCP, function-calling):

PRXY_PIPE=mcp-optimizer,semantic-cache,patterns,ipc

See Customize a pipeline for the full reference.

Streaming

streamText works exactly the same. Cache hits are replayed as synthetic SSE so the streaming UI doesn’t notice. Cache misses pass through the provider stream untouched.

import { anthropic } from '@ai-sdk/anthropic';
import { streamText } from 'ai';
 
const result = streamText({
  model: anthropic('claude-sonnet-4-6'),
  prompt: 'Tell me a story',
});
 
for await (const chunk of result.textStream) {
  process.stdout.write(chunk);
}

Common issues

• Headers being overwritten? The AI SDK passes through your apiKey as Authorization: Bearer … — that’s exactly what prxy.monster expects. Don’t add a custom Authorization header on top.
• Using tools parameter? The mcp-optimizer module sees those tool definitions and prunes the irrelevant ones. If you only want a few tools loaded, you don’t need to do anything — the optimizer handles it.
• generateObject / streamObject? Both work. Schema-based structured output is just a tool call under the hood, and prxy.monster passes it through.

Full example

Working Next.js 15 chat app: github.com/Ekkos-Technologies-Inc/prxy-monster-examples/tree/main/examples/nextjs-vercel-ai