AgentKit by InngestTypeScript 框架，用于构建和编排带工具和工作流的 AI 代理。

4.8 (4)

审阅者 Daniel Nikulshyn·更新 2026年7月

Open Source TypeScript AI Agents Developer Tools Orchestration Workflow Automation Framework

概览

AgentKit by Inngest 是一个开源的 TypeScript 框架，用于构建 AI 代理系统，范围从简单的单模型调用到复杂的多代理编排。它提供了用于定义代理、为其配备工具以及在代理之间路由任务的基础组件，并在各步骤之间保持状态。由 Inngest 背后的团队打造，这个框架与他们的持久化执行平台集成，能够处理长期运行的代理工作流、重试和可观测性。开发者可以连接多个 LLM 提供商，将外部 API 暴露为工具，并组合专门的代理网络，协同解决任务。它面向希望在无需编写自定义编排逻辑的情况下，为 agentic 应用提供生产级可靠性的工程师，提供代码优先的方式，而非可视化构建器。

主要功能

代理与网络原语
带结构化输入的工具调用
有状态多代理路由
多 LLM 提供商支持
通过 Inngest 的持久工作流执行
可观测性和步骤级重试

价格

模型: Freemium
分类: 代码助手
评分: 4.8 / 5 (4)

使用场景

构建多代理研究系统

组建专门化代理网络，在它们之间路由任务，以研究主题、总结发现并生成结构化输出，同时在各步骤间保持状态。

生产级 AI 工作流

使用 Inngest 的持久执行，运行长时间运行的代理工作流，支持自动重试、可观测性和步骤级恢复，确保可靠的生产部署。

工具增强的 LLM 应用

为代理配备结构化工具调用到外部 API，使其能够在 TypeScript 应用中查询数据库、触发动作或获取实时数据。

多提供商 LLM 编排

构建利用多 LLM 提供商的代理系统，将任务路由到最合适的模型，并组合单模型调用或复杂的多代理协作。

优点 & 缺点

优点

开源且原生 TypeScript
支持多代理编排和路由
与 Inngest 集成以实现持久执行
灵活的工具和模型提供商支持

缺点

仅限 TypeScript/JavaScript 生态
需要编程知识，缺乏可视化构建器
关键功能绑定于 Inngest 平台

评测

4.8

4 个评分的平均值。

登录以留下评测。

Tariq Aziz

May 13, 2026

Does the job

Pretty happy overall. Observability and step-level retries just works and open source and TypeScript-native. but no dealbreakers — I'd recommend it to a friend without hesitating.

Beatriz Costa

Feb 17, 2026

Skeptical, then convinced

I went in skeptical — most tools in this space overpromise. It actually delivers on agent and network primitives, and flexible tool and model provider support caught me off guard. Best features tied to Inngest platform is why this isn't a perfect score, still, I'd recommend giving it a real trial.

Aisha Khan

Dec 1, 2025

Skeptical, then convinced

I went in skeptical — most tools in this space overpromise. It actually delivers on multiple LLM provider support, and supports multi-agent orchestration and routing caught me off guard. Limited to TypeScript/JavaScript ecosystem is why this isn't a perfect score, still, I'd recommend giving it a real trial.

Liam O’Connor

Nov 22, 2025

Skeptical, then convinced

I went in skeptical — most tools in this space overpromise. It actually delivers on multiple LLM provider support, and integrates with Inngest for durable execution caught me off guard. Best features tied to Inngest platform is why this isn't a perfect score, still, I'd recommend giving it a real trial.

问答

Do I need a visual builder to create multi‑agent networks?

No. AgentKit is code‑first; you define agents, tools, and routing logic programmatically using its primitives, so a visual builder is not required.

Asked by Ivana Novak · Sep 16, 2025

Which LLM providers can I connect to?

AgentKit supports OpenAI, Anthropic, Gemini, and any model that follows the OpenAI API compatibility layer, letting you switch providers without changing your code.

Asked by Sanjay Gupta · Aug 31, 2025

Can I run AgentKit workflows on my own infrastructure?

Yes. AgentKit is open source and can be run locally, but its durable execution, retries, and observability features are tied to the Inngest platform, which you can integrate with or replace if you handle orchestration yourself.

Asked by Boris Yankov · Aug 13, 2025

What programming languages does AgentKit support?

AgentKit is a TypeScript‑native framework, so it can only be used within the TypeScript/JavaScript ecosystem.

Asked by Miriam Cohen · Aug 2, 2025