Platform Configuration

Configure your LangGraph Platform deployment

Configuration File

The langgraph.json file in your project root defines your deployment configuration:

langgraph.json

{
  "dependencies": ["."],
  "graphs": {
    "agent": "./langgraph/agent.py:graph"
  },
  "env": {
    "ANTHROPIC_API_KEY": "",
    "OPENAI_API_KEY": "",
    "LANGSMITH_TRACING": "true",
    "LANGSMITH_PROJECT": "mcp-server-langgraph"
  },
  "python_version": "3.11"
}

Configuration Options

dependencies · array · required

List of dependency sources. Use ["."] to include current directory, or specify paths to local packages.

Example: "dependencies": [".", "../shared-libs"]

graphs · object · required

Map of graph names to their module paths. Format: "name": "path/to/file.py:variable"

Example: "graphs": { "agent": "./langgraph/agent.py:graph", "assistant": "./langgraph/assistant.py:assistant_graph" }

The variable name (after :) must be a compiled LangGraph graph.

env · object

Environment variables for your deployment. Never put actual API keys here - use LangSmith secrets instead.

Example: "env": { "MODEL_NAME": "claude-sonnet-4-5-20250929", "LANGSMITH_TRACING": "true" }

python_version · string

Python version: "3.10", "3.11", or "3.12". Default: "3.11"

dockerfile_lines · array

Custom Dockerfile commands to run during build.

Example: "dockerfile_lines": ["RUN apt-get update && apt-get install -y curl", "RUN uv pip install custom-package"]

Graph Definition

Your graph must be defined in a Python module and exported as a variable:

langgraph/agent.py

from langgraph.graph import StateGraph, END
from langgraph.checkpoint.memory import MemorySaver

## Define your graph
workflow = StateGraph(...)
workflow.add_node("agent", agent_node)
workflow.set_entry_point("agent")
workflow.add_edge("agent", END)

## IMPORTANT: Export as 'graph' variable
graph = workflow.compile(checkpointer=MemorySaver())

The variable name in langgraph.json must match the exported variable name in your Python file.

Dependencies

Dependencies are specified in langgraph/requirements.txt:

langgraph/requirements.txt

langgraph>=0.2.0
langchain-core>=0.3.0
langchain-anthropic>=0.2.0
langchain-openai>=0.2.0
langsmith>=0.1.0

Keep dependencies minimal for faster builds and lower cold start times.

Environment Variables

Setting Variables

Set environment variables in two ways:

1. langgraph.json - For non-sensitive config
2. LangSmith Secrets - For API keys and sensitive data

## Set secrets (recommended for API keys)
langsmith secret set ANTHROPIC_API_KEY "your-key"
langsmith secret set OPENAI_API_KEY "your-key"

Available at Runtime

Access environment variables in your code:

import os

api_key = os.environ["ANTHROPIC_API_KEY"]
model_name = os.environ.get("MODEL_NAME", "claude-sonnet-4-5-20250929")

Multiple Environments

Create separate deployments for each environment:

Development

# Deploy to dev
langgraph deploy my-agent-dev --tag development

# Use dev-specific secrets
langsmith secret set ANTHROPIC_API_KEY "dev-key" --deployment my-agent-dev

Staging

# Deploy to staging
langgraph deploy my-agent-staging --tag staging

# Use staging-specific LangSmith project
LANGSMITH_PROJECT=my-agent-staging langgraph deploy my-agent-staging

Production

# Deploy to production
langgraph deploy my-agent-prod --tag production

# Use production secrets
langsmith secret set ANTHROPIC_API_KEY "prod-key" --deployment my-agent-prod

Advanced Configuration

Custom Build Steps

Add custom build commands:

{
  "dockerfile_lines": [
    "RUN apt-get update && apt-get install -y build-essential",
    "RUN uv pip install --no-cache-dir torch torchvision"
  ]
}

Multiple Graphs

Deploy multiple graphs in one deployment:

{
  "graphs": {
    "chat": "./langgraph/chat.py:chat_graph",
    "analysis": "./langgraph/analysis.py:analysis_graph",
    "summary": "./langgraph/summary.py:summary_graph"
  }
}
```python
Access via different endpoints:
- `POST /chat/invoke`
- `POST /analysis/invoke`
- `POST /summary/invoke`

### Validation

Validate your configuration before deploying:

Test locally

langgraph dev

Verify graph can be imported

python -c “from langgraph.agent import graph; print(graph)“

### Next Steps

<CardGroup cols={2}>
  <Card title="Secrets Management" icon="key" href="/deployment/platform/secrets">
    Manage API keys securely
  </Card>
  <Card title="Deploy Now" icon="rocket" href="/deployment/platform/quickstart">
    Deploy your configured agent
  </Card>
  <Card title="Monitoring" icon="chart-line" href="/deployment/platform/monitoring">
    Monitor your deployment
  </Card>
  <Card title="CI/CD" icon="infinity" href="/deployment/platform/ci-cd">
    Automate deployments
  </Card>
</CardGroup>

### Reference

#### Example langgraph.json

Complete configuration file with all options:

```json
{
  "dependencies": ["."],
  "graphs": {
    "agent": "./langgraph/agent.py:graph"
  },
  "env": {
    "MODEL_NAME": "claude-sonnet-4-5-20250929",
    "LANGSMITH_TRACING": "true",
    "LANGSMITH_PROJECT": "my-agent-prod",
    "MAX_ITERATIONS": "10"
  },
  "python_version": "3.11",
  "dockerfile_lines": []
}

Example graph export

How to export a compiled graph for LangGraph Platform:

from langgraph.graph import StateGraph, END
from langgraph.checkpoint.memory import MemorySaver
from typing import TypedDict, Annotated

class State(TypedDict):
    messages: Annotated[list, "The conversation messages"]

def agent_node(state: State):
    # Your agent logic
    return state

workflow = StateGraph(State)
workflow.add_node("agent", agent_node)
workflow.set_entry_point("agent")
workflow.add_edge("agent", END)

# Export for platform
graph = workflow.compile(checkpointer=MemorySaver())