LangGraph Examples

LangGraph is a library for building stateful, multi-step AI agents as graphs of computation. AgentVisor™ integrates natively with LangGraph, providing durable execution via Temporal, policy-enforced HTTP proxying, and seamless checkpointing — so your graph-based agents run safely without changes to your application code.

Prerequisites

• Ollama (or another LLM endpoint) — most examples use ollama via Docker Compose
• Temporal — required for durable execution and checkpointing
• AgentVisor CLI — agentvisor serve . starts the runtime
• jq — used by the examples below to pretty-print API responses

The included compose.yml in each template starts both Ollama and Temporal locally.

Examples

Work through these in order to build from fundamentals to expert patterns:

Step	Example	Difficulty	Key Feature
1	Chatbot Agent	Basic	add_messages, checkpointing, HTTP proxy
2	QA Agent	Intermediate	bind_tools(), ReAct pattern, tool policies
3	Task Agent	Intermediate	interrupt(), human-in-the-loop, durable approval
4	MCP Agent	Intermediate	MCP Gateway, dynamic tool discovery
5	Tracing Agent	Intermediate	OpenLLMetry, manual spans, Jaeger, agent span forwarding
6	Research Agent	Advanced	Fine-grained HTTP policies, internal network blocking
7	Coordinator Agent	Expert	Multi-agent orchestration, sub-agent policies
8	Pentest Agent	Expert	gVisor shell execution, credential substitution, agentvisor build

How It Works

LangGraph agents in AgentVisor work through the LangGraph framework provider:

1. Discovery: AgentVisor detects langgraph.json and selects the LangGraph provider automatically — checked first, ahead of every other framework's discovery file
2. Schema extraction: Graph schemas are extracted via a three-tier fallback — a build-time cache, then dynamic extraction, then graph names only from langgraph.json
3. Execution: python3 -m agentvisor.langgraph.runner loads and invokes your compiled graph inside the sandbox
4. Checkpointing: AgentVisorCheckpointer is a drop-in BaseCheckpointSaver that stores graph state in Temporal automatically

Key Concepts

Concept	What It Is	Where You'll See It
StateGraph	LangGraph class for defining agent flow as a directed graph	All examples
add_messages	Annotation that automatically accumulates conversation history in state	Chatbot, QA, Task
bind_tools()	Attaches tool definitions to an LLM so it can call them	QA, Research, Coordinator
interrupt()	Pauses graph execution and waits for human input via the Thread API	Task Agent
langgraph.json	Config file AgentVisor reads to discover your graph entry points	All examples
AgentVisorCheckpointer	Drop-in BaseCheckpointSaver that stores state in Temporal	Chatbot, QA, Task, Coordinator

Quick Start

# Scaffold any example from a template
agentvisor template create langgraph/chatbot-agent
cd chatbot-agent

# Start Temporal and Ollama
docker compose up -d

# Run the agent
agentvisor serve .

Then open http://localhost:8090/swagger-ui/ to explore the API.

See Also

• LangGraph Agents Guide — graph discovery, schema extraction, and framework configuration in depth