Web search
Ground answers in live web sources via the
web_search_preview tool. The model issues search queries, reads snippets, and returns an answer with citations — no orchestration on your side.OpenAI
Pair
openaiClientAdapter with openaiResponsesRunAdapter:import OpenAI from 'openai'
import {
supercompat,
openaiClientAdapter,
openaiResponsesRunAdapter,
memoryStorageAdapter,
} from 'supercompat/openai'
const client = supercompat({
clientAdapter: openaiClientAdapter({ openai: new OpenAI() }),
storageAdapter: memoryStorageAdapter(),
runAdapter: openaiResponsesRunAdapter(),
})
const response = await client.responses.create({
model: 'gpt-4.1',
input: 'What is the current population of Tokyo? Use web search.',
tools: [{ type: 'web_search_preview' }],
})
The returned text includes inline citations — each citation is a
url_citation annotation on the text output item.Reading citations
for (const item of response.output) {
if (item.type !== 'message') continue
for (const content of item.content) {
if (content.type !== 'output_text') continue
for (const annotation of content.annotations ?? []) {
if (annotation.type === 'url_citation') {
console.log(annotation.url, annotation.title)
}
}
}
}
Azure OpenAI
Same tool declaration against an Azure OpenAI Responses deployment:
import { AzureOpenAI } from 'openai'
import {
supercompat,
azureOpenaiClientAdapter,
azureResponsesRunAdapter,
memoryStorageAdapter,
} from 'supercompat/openai'
const azureOpenai = new AzureOpenAI({
endpoint: process.env.AZURE_OPENAI_ENDPOINT!,
apiKey: process.env.AZURE_OPENAI_API_KEY,
apiVersion: '2024-10-21',
})
const client = supercompat({
clientAdapter: azureOpenaiClientAdapter({ azureOpenai }),
storageAdapter: memoryStorageAdapter(),
runAdapter: azureResponsesRunAdapter(),
})
const response = await client.responses.create({
model: 'my-gpt-4-1-deployment', // Azure deployment name
input: 'What is trending in AI this week?',
tools: [{ type: 'web_search_preview' }],
})
Anthropic
Anthropic exposes a native server-side web-search tool,
web_search_20250305. The anthropicClientAdapter forwards it as-is and normalizes the resulting web_search_tool_result content blocks for you. Pair with completionsRunAdapter on the Assistants surface:import Anthropic from '@anthropic-ai/sdk'
import { PrismaClient } from '@prisma/client'
import {
supercompat,
anthropicClientAdapter,
completionsRunAdapter,
prismaStorageAdapter,
} from 'supercompat/openai'
const client = supercompat({
clientAdapter: anthropicClientAdapter({ anthropic: new Anthropic() }),
storageAdapter: prismaStorageAdapter({ prisma: new PrismaClient() }),
runAdapter: completionsRunAdapter(),
})
const assistant = await client.beta.assistants.create({
model: 'claude-sonnet-4-6',
instructions: 'Search the web before answering.',
tools: [
{
type: 'web_search_20250305',
web_search_20250305: {
max_uses: 5,
},
} as any,
],
})
const thread = await client.beta.threads.create()
await client.beta.threads.messages.create(thread.id, {
role: 'user',
content: 'What is trending in open-source AI today?',
})
const run = await client.beta.threads.runs.createAndPoll(thread.id, {
assistant_id: assistant.id,
})
The tool executes on Anthropic's side — your run completes without
requires_action, and the results appear in the message content.Perplexity (web-grounded by default)
Perplexity's Sonar models ground every response in live web sources without any
tools entry. Pair perplexityClientAdapter with completionsRunAdapter:import OpenAI from 'openai'
import {
supercompat,
perplexityClientAdapter,
completionsRunAdapter,
memoryStorageAdapter,
} from 'supercompat/openai'
const perplexity = new OpenAI({
apiKey: process.env.PERPLEXITY_API_KEY,
baseURL: 'https://api.perplexity.ai',
})
const client = supercompat({
clientAdapter: perplexityClientAdapter({ perplexity }),
storageAdapter: memoryStorageAdapter(),
runAdapter: completionsRunAdapter(),
})
const response = await client.responses.create({
model: 'sonar-pro',
input: 'What happened in open-source AI this week?',
})
No
tools configuration needed — Sonar always searches.Compatibility
For providers without a built-in web-search tool (Google, Mistral, Groq, Together, OpenRouter, Ollama), build a function tool that hits your own search backend.