Practical overview
When it fits
Use it when building a TypeScript application that requires coordinated agents, deterministic routing or MCP tooling.
Who it is for
- TypeScript developers building applications that coordinate multiple agents.
- Teams designing agent networks that need MCP-enabled tools.
Documented uses
- Coordinate multiple agents within a TypeScript application.
- Route work deterministically between agents.
- Connect an agent network to tools through MCP.
Requirements & limits
- The project is a framework for building agent networks, not a finished agent application.
- Its stated focus combines deterministic routing with MCP tooling.
- The repository specifies an Apache-2.0 license.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Create and test a single AgentKit agent
Set up a TypeScript project with one agent, run its server, and invoke it through the Inngest dev server.
Before you start
- An existing project, or a new directory initialized with npm, pnpm, or yarn.
- AgentKit and Inngest installed together.
- An Anthropic API key set as an environment variable.
- The documented quick-start uses
tsxand the Inngest CLI for local execution and testing.
Steps
Install
@inngest/agent-kitandinngestwith your package manager; for a new project, create a directory and initialize it first.In
index.ts, define an agent withcreateAgent, including its name, description, system prompt, and model. SetANTHROPIC_API_KEYas an environment variable.Create a server with
createServer({ agents: [agent] })and callserver.listen(3000, ...).Run the file with the command below, then start the Inngest dev server pointing to
http://localhost:3000/api/inngestusing the command below.npx tsx ./index.tsnpx inngest-cli@latest dev -u http://localhost:3000/api/inngestOpen the dev server’s functions page, invoke the agent function with an input prompt, and inspect the run output.
Expected outcome
A locally running AgentKit server with a single agent that can be invoked and inspected through the Inngest dev server.
Guide limitations
- This recipe follows the documented Anthropic model example and requires an Anthropic API key.
- The guide documents local testing, not production deployment.
- The quick-start says
inngestis a required peer dependency starting with AgentKit v0.9.0; the installation page says v0.8.0, so check the package version’s current requirements.
Build and test a two-agent network
Combine database administration and security agents in a network with state-based routing, then invoke the network locally.
Before you start
- A working AgentKit setup with the agents and local Inngest dev server from the quick-start.
- Two agents configured with their respective roles and model settings.
- A shared network state and a router that selects agents.
Steps
Add a
save_answertool to each agent that writes its answer into the shared network state under a distinct key.Create a network with
createNetwork, include both agents, define the shared state type, and configure a router that calls the security agent first, the database administrator next, and then ends the network.Register both agents and the network with
createServer.With the Inngest dev server running, invoke the network function from its functions page using the documented JSON input format, then inspect the finalization step’s JSON payload.
Expected outcome
A locally testable network that routes between two specialized agents and stores their answers in shared state.
Guide limitations
- The documented example uses Anthropic’s Claude model; provider configuration and credentials are needed for execution.
- The example’s router follows a fixed state-based sequence; the cited steps do not document reasoning-based routing.
- The guide documents local testing, not production deployment.
Compare options for Agent orchestration
These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.
| Project | Documented purpose | Platform | Access and requirements |
|---|---|---|---|
| AgentKit | Coordinate multiple agents within a TypeScript application. | Linux, macOS, Windows | |
| Agenite | Define typed tools with JSON Schema validation and error handling. | CLI | |
| Veldra | Create an agent with a policy, tools, skills, knowledge base and reasoning method. | Self-hosted | The documented deployment is self-hosted and uses a sign-in gate after a first-run administrator account is created. |
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 15 October 2026
- Original website title
- inngest/agent-kit · Official homepage title · retained in its source language

