Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.langchain.com/llms.txt

Use this file to discover all available pages before exploring further.

This guide builds a data analysis agent from first principles using create_agent and deepagents middleware. Rather than starting with create_deep_agent, we assemble the harness one piece at a time: so you can see exactly what each component adds and swap in only what your use case needs. The agent we’ll build:
  1. Accepts a CSV file for analysis
  2. Writes and executes Python code in an isolated sandbox
  3. Delegates visualization work to a specialized subagent
  4. Loads data analysis patterns from a skills file

​
Setup

Enable LangSmith tracing to inspect every step: 
export LANGSMITH_TRACING=true
export LANGSMITH_API_KEY=...

​
Step 1: The minimal agent

A model, a loop. Nothing else yet. This runs, but the agent has no filesystem and no way to execute code. The next steps add those.

​
Step 2: Add a sandbox backend

LangSmithSandbox gives the agent an isolated environment with a filesystem and an execute tool for running shell commands. The agent can install packages, write scripts, and run them: without touching the host. FilesystemMiddleware adds read_file, write_file, edit_file, glob, and grep. Because LangSmithSandbox implements the sandbox protocol, it also adds execute: the agent can now run shell commands. Upload a CSV and invoke:

​
Step 3: Add context management

For longer analysis sessions the context window fills. SummarizationMiddleware compresses history automatically so the agent keeps working without hitting token limits.

​
Step 4: Add skills

Skills give the agent on-demand domain knowledge via progressive disclosure: loaded only when the current task calls for it. Create a skill file in your skills directory: 
skills/
  pandas-patterns/
    SKILL.md

---
name: pandas-patterns
description: Common pandas and matplotlib patterns for data analysis and visualization
---

## Data loading
Use `pd.read_csv()` for CSV files. Always check `df.info()` and `df.describe()` first.

## Visualization
Use `matplotlib` for bar charts, `seaborn` for statistical plots.
Save figures with `plt.savefig("output.png", dpi=150, bbox_inches="tight")`.

## Reporting
Write a markdown summary to `report.md` alongside any generated charts.

​
Step 5: Add a visualization subagent

Some tasks benefit from isolation. A visualization subagent runs in its own context window, keeping chart generation separate from the main analysis: and enabling parallel execution. The main agent handles analysis and planning; it delegates chart generation to the visualizer subagent via the task tool.

​
What you built

MiddlewareWhat it adds
FilesystemMiddleware + LangSmithSandboxIsolated filesystem + execute tool
SummarizationMiddlewareAutomatic context compression
SkillsMiddlewareDomain knowledge loaded on demand
TodoListMiddleware + SubAgentMiddlewareParallel visualization subagent
This is the same foundation as create_deep_agent: assembled manually so you control exactly what’s included. The possibilities don’t end here: see Prebuilt middleware for the full list of composable capabilities, and the create_agent reference for all configuration options.
Connect these docs to Claude, VSCode, and more via MCP for real-time answers.
Edit this page on GitHub or file an issue.