AgentKit

AgentKit is for developers who know their tools and workflow well but do not want to build the agent's sessions, tool-call chains, and message flow from scratch.

What it solves

It helps with:

• agent session initialization
• streaming messages
• tool calls
• tool callbacks
• session history, stop, and cleanup

Best fit

• you have tools but do not want to assemble the agent yourself
• you want to validate an agent workflow quickly
• you want the platform to own the orchestration layer

Core capabilities

Session initialization

Common entry points include:

• initSession
• initSimple

Conversation execution

Common patterns include:

• chat: a streaming, session-based request
• chatSimple: a one-shot lightweight request

Tool callbacks

When the agent triggers a tool call, the application can send results back through:

• callback
• streamCallback

When AgentKit is a better fit than ModelKit

• you do not just need single-step model calls
• you need “model + tools + session” orchestration together
• you do not want to maintain the agent state machine yourself

When you may not need AgentKit

• you already have your own complex agent runtime
• you are only building single-step Q&A, summarization, translation, or classification flows

Next steps

• Read Choose Your Stack
• Read TypeScript Quickstart
• Read the AgentKit route descriptions in agentos-api README