Production-ready Toolkit
for AI Agents

Build intelligent agents with ease. Powerful SDK with pre-built components, seamless integrations, and production-ready tools.

Core Framework

Minimal abstractions. Maximum power.

main.go
AgentKit
1package main
2
3import "github.com/model-box/agent-kit"
4
5func main() {
6agent := NewAgent()
🤖Core AI logic
7 runner := NewRunner()
Execution engine
8 runner.Start(agent)
🚀Deploy & run
9}
BUILDING BLOCKS
🤖
Agent
Core logic
🔧
Tool
Actions
Runner
Executor
🛡️
Guard
Safety

Minimal Abstractions

Simple by design

Start with just few lines of code. As you grow, Agent, Tool, Handoff, Guardrail, and Runner provide a complete foundation for any AI system you can imagine.

Provider Agnostic

20+ model vendors

Unified interface covering OpenAI, Anthropic, Google, local models, and all major modalities. Switch providers with a single line change using ModelBox bridge.

1
Analyze task
2
Generate plan
3
Execute action
4
Evaluate result
25%
Processing: Analyze task

Auto-Iteration

Built-in agent loop

Agents automatically iterate until tasks are complete. Just define the goal and let the framework handle execution.

🔍
Research
✍️
Writer
👀
Review
🚀
Deploy
🔍
Research Agent
Data analysis

Multi-Agent

Dynamic handoffs

Seamless delegation to specialized sub-agents. Build complex workflows spanning single agents, teams, and hierarchical agent organizations.

Agent Monitor

Ultra-lightweight runtime

READY
Startup Time
847μs
Memory Usage
2.4KB
CPU Usage
0.1%
$ agentkit start
⚡ Initializing runtime...
🔧 Loading agents...
✓ Ready in 847μs
OPTIMIZED

Ultra-Light

Microsecond startup

Kilobyte-scale memory footprint with lightning-fast initialization. Perfect for large scale and serverless deployments environments.

Production Ready

Enterprise-grade safety and observability.

Native Tracing

Real-time monitoring

Built-in tracing that streams logs to OpenAI and ModelBox Dashboard. Track agent reasoning, tool usage, and performance metrics in real-time.

OpenAI Native

First-class integration

Official services integration: search, retrieval, voice, and more.

Guardrails

Type-safe validation

Input validation and type-safe outputs with built-in guardrail layer.

Memory & RAG

Built-in intelligence

Memory store, session management, and agentic RAG toolkit. Rich reasoning utilities with chain-of-thought orchestration for complex problem solving.