此语言仍有 24 条内容或界面任务待翻译;未完成部分显示中文原文。
智能体与自动化 · 火山引擎

AgentKit API

Call the Volcano Engine AgentKit agent API and get a conversational agent with tool calling and memory running in 30 minutes.

AI 辅助整理
agent developmenttool callingAPIconversational appsstreaming response

Tool Introduction

AgentKit is Volcano Engine's API service for agent development. Input is a POST request containing agent_id, query, and session_id, and output is a JSON response with an answer field. It ships with built-in tools such as weather, search, and calculator, supports custom tools and streaming responses, and provides session memory. It is well suited for conversational agents with high daily call volumes that need tool calling and memory. For scenarios that only require a single LLM inference pass, using the LLM directly with API may be more cost-effective.

Who It's For

Conversational agent developers who need built-in tool calling/memory capabilities and a daily call volume above 1 ten thousand API; small and medium development teams that don't want to build an underlying agent framework from scratch and want to launch quickly; enterprise-level agent development scenarios requiring multi-model orchestration and custom workflows. Not suitable for scenarios that only require single-pass inference with no tool calling/memory needs, or secondary developers who need full control over the underlying framework source code.

Recommended Steps

  1. 01

    Prepare a Python 3.8+ or Node.js 16+ environment, complete real-name verification of a Volcano Engine account, activate AgentKit service permissions, and obtain AK/SK keys.

  2. 02

    Install the official SDK: pip install volcengine-agentkit==1.2.0.

  3. 03

    Set the AK/SK and region as environment variables; supported regions are currently cn-beijing and cn-shanghai.

  4. 04

    Initialize AgentKitClient with an agent ID created in the console, and construct a request containing agent_id, query, and session_id to call the run_agent API and verify basic connectivity.

  5. 05

    Log in to the console and go to the agent configuration page; in the tool management module, enable tool switches such as web search, save, and re-run the call to verify the tools take effect.

  6. 06

    Go to the console's call statistics page to check the latency, response status, and token consumption of each request for troubleshooting and cost tracking.

Know Before You Use

Requires a Volcano Engine account with completed real-name verification and AgentKit service permissions activated, plus AK/SK keys; the region must have AgentKit service enabled and match the region parameter in the code, otherwise 403 is returned. All AgentKit APIs support only POST requests, not GET. Supported regions are currently cn-beijing and cn-shanghai. For real-time output, add stream:true to the request parameters to enable streaming responses. The specific conditions of the free trial version should be verified on the official website.

费用与套餐:Costs consist of token consumption fees plus tool call fees: basic conversation calls cost 0.01 yuan per 1000 tokens, and each tool call costs 0.005 yuan (based on the pricing page data cited in the article as of 8 2026). For low-traffic test scenarios with fewer than 100 calls per day, the article recommends prioritizing the free trial version or a lightweight SDK. Pricing is subject to the official website and is not guaranteed to be permanent.

Product Selection Assistant

让需求变成可修改的筛选条件

依据已录入参数生成条件,结果可查看和修改。帧率与性能需核对实测,不能仅凭算力数值保证。