Web search

Overview

Tool identifier web_search. web_search lets the agent run web-wide searches. By default it uses the Volcengine Unified Information Search API; see the API docs. When CLOUD_PROVIDER is set to byteplus, the tool switches to the BytePlus SearchInfinity API. Import path: from veadk.tools.builtin_tools.web_search import web_search parallel_web_search provides parallel multi-query search and likewise uses the BytePlus SearchInfinity API in BytePlus mode.

Environment & prerequisites

Requirements:

Configure Volcengine AK / SK, or use a temporary StsToken issued via Volcengine IAM. In BytePlus mode, use BYTEPLUS_WEB_SEARCH_API_KEY instead.
Configure the API key for the agent’s reasoning model.

Environment variable	Mode	Default	Description
`MODEL_AGENT_API_KEY`	All	—	API key for the agent’s reasoning model.
`CLOUD_PROVIDER`	All	`volcengine`	Cloud provider; set to `byteplus` to switch to BytePlus search.
`VOLCENGINE_ACCESS_KEY`	Volcengine	—	Volcengine AccessKey.
`VOLCENGINE_SECRET_KEY`	Volcengine	—	Volcengine SecretKey.
`BYTEPLUS_WEB_SEARCH_API_KEY`	BytePlus	—	API key for the BytePlus SearchInfinity API; required in BytePlus mode.
`BYTEPLUS_WEB_SEARCH_URL`	BytePlus	`https://torchlight.byteintlapi.com/search_api/web_search`	Override the BytePlus SearchInfinity search endpoint.

Or configure in config.yaml:

config.yaml

model:
  agent:
    provider: openai
    name: doubao-seed-2-1-pro-260628
    api_base: https://ark.cn-beijing.volces.com/api/v3/
    api_key: your-api-key-here
volcengine:
  # used by Viking DB, web_search, and other tools
  access_key: your-access-key-here
  secret_key: your-secret-key-here

BytePlus mode example:

export CLOUD_PROVIDER="byteplus"
export BYTEPLUS_WEB_SEARCH_API_KEY="your-byteplus-search-api-key"
export MODEL_AGENT_API_KEY="your-model-api-key"

Usage

examples/tools/web_search/agent.py

import asyncio

from veadk import Agent, Runner
from veadk.memory.short_term_memory import ShortTermMemory
from veadk.tools.builtin_tools.web_search import web_search

agent = Agent(
    name="web_search_agent",
    model_name="doubao-seed-2-1-pro-260628",
    description="An agent that answers questions with web search.",
    instruction="You are a helpful assistant. Use the web_search tool when you need fresh information.",
    tools=[web_search],
)

runner = Runner(agent=agent, short_term_memory=ShortTermMemory())


async def main():
    response = await runner.run("What's the weather in Hangzhou today?")
    print(response)


if __name__ == "__main__":
    asyncio.run(main())

Parameters and results

Tool	Parameter	Type	Default	Result
`web_search`	`query`	`str`	Required	Summary list for one query
`parallel_web_search`	`queries`	`list[str]`	Required	Mapping from query strings to summary lists
Both	`tool_context`	`ToolContext \| None`	`None`	Injected during agent calls; may supply Volcengine credentials

In Volcengine mode, credentials are resolved in this order: paired TOOL_WEB_SEARCH_ACCESS_KEY / TOOL_WEB_SEARCH_SECRET_KEY, AK/SK in the tool context, VOLCENGINE_ACCESS_KEY / VOLCENGINE_SECRET_KEY, and IAM temporary credentials from a managed environment. The session token comes from that final IAM flow; setting AK/SK directly does not attach an arbitrary STS token. Each query requests up to five results. Parallel results use query text as dictionary keys, so deduplicate queries first. An error response may appear as a list item; inspect results before using them in an answer. This example calls the search service without a reasoning model:

import asyncio
from veadk.tools.builtin_tools.parallel_web_search import parallel_web_search

async def main():
    results = await parallel_web_search(["AgentKit documentation", "VeADK documentation"])
    for query, summaries in results.items():
        print(query, summaries)

asyncio.run(main())

Overview

Get started

Core components

Observability

Extensions

Frontend

Security

Deployment

Overview

Environment & prerequisites

Usage

Parameters and results

​Overview

​Environment & prerequisites

​Usage

​Parameters and results

Overview

Environment & prerequisites

Usage

Parameters and results