Custom Demo Agent

Summary

This repo is a chat agent built on deepagents. It showcases every feature of DeepAgents (and many features of LangSmith).

• DeepAgents: Subagents, Skills, Code-execution, Generative UI, file read/write, voice, ...
• LangSmith: Context Hub, deployments, evals, sandboxes, monitoring, Engine, ...

If you're wanting to build your own chat agent, this should hopefully serve as a useful reference for your coding agent to see how each feature is used.

Customizable!

Customize agents for your use case. Customize the:

• system prompt
• skills
• tools (through suppling your own mcp server urls)
• subagents (through suplying your own A2A agent urls)
• UI "skin" (logo/name/color)

If you're a LangChain employee, you can access a hosted version here. It requires a password. Ask @josiahcoad for it. Otherwise you can easily run locally and just supply your own LangSmith Api Key.

Run locally

cp .env.example .env

Set LANGSMITH_API_KEY and ANTHROPIC_API_KEY in the root .env.

uv sync --group dev
uv run python scripts/preflight.py     # checks connectivity; makes real API calls
./run.sh                              # backend :2024, frontend :3000

Open http://127.0.0.1:3000.

Creating your agent in the UI

When you open the UI, you'll be met with a setup modal.

All you need to provide to create a new agent is:

• Company Name (optional)
• Company Website (optional... for setting up the UI Skin)
• Use Case (optional; can use the brand website to infer a use case)

A new use-case/agent takes about 40 seconds to setup. Behind the scenes, we are...

• pulling the brand details
• writing the system prompt
• provisioning a sandbox
• creating some dummy files in the sandbox
• creating some skills
• creating some "quick prompts" for you to click to demo the agent

Tip: Optionally enable demo traffic to populate 200 sample traces for Monitoring, Insights and Engine.

You can further tweak your agent in the side-panel.

Architecture

A deepagent, using assistants to store per-use-case configuration, plus a React frontend:

• Setup resolves the customer scenario and prepares data, skills, prompts and evals.
• Runtime applies the assistant's model/tools, reads its prompt fresh and runs the agent.
• Sandbox owns working files; Context Hub stores prompts and skills.
• Frontend streams tool activity, dashboards and HTML assets from the same conversation.

Implementation details: AGENTS.md. Development and checks: CLAUDE.md. More demos: voice, MCP Apps, release evals (these score the planted bug firing, opposite to presenter evals).