Documentation
Welcome to the AgentSea documentation. Learn how to build powerful agentic AI applications.
Documentation Sections
Quick Start
Get up and running with AgentSea in minutes
Agentic Coding
Interactive AI coding sessions with 13 built-in tools
Agents
Learn how to create and configure intelligent agents
Multi-Agent Crews
Role-based coordination with delegation strategies
Per-Model Type Safety
Compile-time validation for model-specific options
Local & Open Source
Run agents locally for privacy and zero API costs
Tools
Extend agent capabilities with built-in and custom tools
LLM Gateway
OpenAI-compatible API with intelligent routing
Guardrails
Content safety, validation, and prompt injection protection
LLM Evaluation
Metrics, LLM-as-Judge, and human feedback
Red Teaming
Adversarial testing, vulnerability scanning, and compliance
Embeddings
Multi-provider embeddings with caching and vector stores
Semantic Cache
Exact-match, semantic, and hybrid LLM response caching
Document Ingestion
PDF, DOCX, HTML & Markdown parsing with smart chunking
Prompt Management
Version control, A/B testing, and environment promotion
Cost Tracking
60+ model pricing registry with budget enforcement
Conversation Analytics
Intent classification, sentiment, and flow analysis
Agent Debugger
Step-through execution, checkpoint replay, and what-if tests
Browser Automation
Computer-use agent with Claude vision
MCP Integration
Connect to MCP servers for external tools
Quick Example
Create your first agent in just a few lines of code:
import { Agent, AnthropicProvider, ToolRegistry } from '@lov3kaizen/agentsea-core';
const agent = new Agent(
{
name: 'my-agent',
model: 'claude-sonnet-4-20250514',
systemPrompt: 'You are a helpful assistant.',
},
new AnthropicProvider(),
new ToolRegistry()
);
const response = await agent.execute('Hello!', context);Key Features
Agentic coding (sea code) with 13 built-in coding tools
Multi-agent crews with 5 delegation strategies
LLM Gateway with intelligent routing and caching
Multi-provider support: 60+ models incl. GPT-5, o3, o4-mini
Guardrails for safety, validation, and security
Red teaming with adversarial testing and compliance checks
LLM evaluation with metrics and LLM-as-Judge
Multi-provider embeddings with vector stores
Semantic caching (exact, similarity, hybrid) with replay
Document ingestion (PDF, DOCX, HTML, Markdown) with chunking
Prompt management with version control and A/B testing
Cost tracking with 60+ model pricing and budget enforcement
Conversation analytics with intent and sentiment tracking
Agent debugger with checkpoint replay and what-if testing
Computer-use & browser automation with Claude vision
Per-model type safety with compile-time validation
Local & open source models (Ollama, LM Studio, vLLM, etc.)
First-class MCP (Model Context Protocol) integration
Built-in voice support (TTS/STT) with multiple providers
Commerce capabilities with ACP protocol (14 tools)
REST API, SSE, and WebSocket support
Advanced memory (Episodic, Semantic, Working)
Enterprise multi-tenancy with API key auth
Comprehensive observability (logging, metrics, tracing)
NestJS integration with decorators and modules
Full TypeScript support with strict types
Ready to get started?
Check out our Quick Start Guide to begin building powerful AI agents.
Get Started →