Deploy AI agents in 5 lines of code.

TL;DR

Build AI-agents in 5 lines of code. Skip the set up & infrastructure. Live and running.

from custodian_labs import Custodian



app = Custodian(

    model="gpt-4o",

    system_prompt="You are a concise, helpful assistant.",

).deploy()



print(app.chat("Hello!").response)

print(app.chat_url)  # ← live, shareable URL

Runnable Colab notebook here.

Skip the infrastructure

DIYing this: FastAPI behind a reverse proxy + Redis for sessions + a small frontend + an auth layer + a vector DB and ingestion pipeline. All of it is annoying. Most teams burn weeks on plumbing before the interesting work even starts.

Setup

Grab a free API key from the dashboard, then:

!pip install -q custodian-labs

import os

from getpass import getpass

os.environ["CUSTODIAN_SDK_API_KEY"] = getpass("API key: ")

The basic deploy

from custodian_labs import Custodian



app = Custodian(

    model="gpt-4o",

    system_prompt="You are a concise, helpful assistant.",

).deploy()



reply = app.chat("In one sentence, what can you help me with?")

print(reply.response)

print("\nShareable chat URL:", app.chat_url)

What you get back:

• 
  app — handle to a live hosted agent.


• 
  reply.response — the text.


• 
  reply.session_id — pass it to the next .chat() to continue the conversation.


• 
  app.chat_url — live URL to a hosted chat UI. Paste in Slack, share with PMs, done.

Adding your own data (RAG)

from custodian_labs import Custodian



agent = Custodian(

    model="gpt-4o",

    system_prompt=(

        "You answer questions about the car dataset you've been given. "

        "Be specific and cite the numbers."

    ),

)

agent.add_data_source_file("data_examples/car_info.csv")

app = agent.deploy()



reply = app.chat(

    "Which SUV has the best city MPG, and what is its annual maintenance cost?"

)

print(reply.response)

print("\nShareable chat URL:", app.chat_url)

add_data_source_file handles chunking, embedding, and retrieval at chat time. Works on CSVs, PDFs, and the usual suspects — calling code doesn't change whether you point it at a CSV or a folder of contracts.

Ideas worth stealing

• 
  Internal "ask the docs" bots — handbook / Notion export / product wiki → drop the chat URL in Slack.


• 
  Customer support assistants — RAG over help-center articles + past tickets, in front of a human escalation path.


• 
  Sales / RFP helpers — past proposals + contracts, so reps can ask "how did we answer security questions for X?"


• 
  Research assistants — folder of PDFs (papers, reports, filings), chat the corpus.


• 
  Onboarding helpers — runbooks + architecture docs so new hires can self-serve without bothering a senior.


• 
  Personal knowledge agents — your notes or saved articles, hosted at a URL you can hit from anywhere.

Same shape for all of them: write a system prompt, attach docs, deploy, share the URL.

Try it

The Colab notebook runs end-to-end with sample data — bring no files, just a free key from the dashboard. Should take ~1 min.

I'd genuinely love feedback — what's missing, what's confusing, what would make you actually use it. Feature requests very welcome. Drop them in the comments, or email [email protected] if you build something with it (or break it in an interesting way).

Custodian Labs SDK

Next posts coming on multi-agent teams with topic routing, and the PII / proprietary-data masking layer for legal and healthcare agents.