From idea tosecure live agentsin 2 minutes.
Custodian Labs takes you from idea to deployed, production-grade AI agent, with privacy built in at every layer. No infrastructure. No data leakage.
Describe your AI
Guardian Layer onFor example: Build me an agent that helps me do my accounting: reconcile invoices and flag anything unusual.
Supported by
Everything you need to ship secure agents.
One platform to build, protect and deploy AI agents, from a single prompt to a production team of agents, on any model.
Detects PII before it reaches any model and removes, masks or transforms it, then relinks the originals in the response. Add your own entity types, from employee names to product codes.
Your prompt
Email James Harrington at james@acme.com re: $240,000
What the model sees
Email Alex Morgan at alex@synthco.io re: $240,000
OpenAI, Anthropic, Google, Mistral, open-source and more. Switch models without touching your agent logic.
Ship with a few lines of Python or JavaScript, or drag the same building blocks together in the visual builder.
from custodian_labs import Custodian
Custodian(model="gpt-4o")
.deploy()Give each agent its own model, prompt and topics. Custodian routes every message to the right one and deploys the team as a single app.
Search the web, read and send Gmail, post to X, scrape pages and parse documents, or plug in any MCP tool.
Upload PDFs, Word docs and spreadsheets. Retrieval, embeddings and citations are handled for you.
1 · refund-policy.pdf, p.2
No vector store to provision, no servers to scale, no retry logic to write. Deploy once and you're live.
You don't run
Your prompt
Email James Harrington at james@acme.com re: $240,000
What the model sees
Email Alex Morgan at alex@synthco.io re: $240,000
Also included
- Python & JS SDKs
- MCP server
- Session memory
- Bring your own key
- Source citations
- Custom PII entities
- Image & document masking
- Usage dashboard
Questions, answered.
What people ask before they ship their first agent, from how we compare to building with other tools, to where your data goes.
Something else? Email us
They help you build an agent for yourself. Custodian puts it to work for your team and customers.
ChatGPT and Claude are great for building an agent for yourself: a custom assistant inside the chat app, or agent code they write for you. Putting that agent in front of your team or customers is a different job. It needs a link or API people outside the app can use, sensitive data kept away from the model, and hosting that keeps it running. Custodian is that production layer, ready on day one.
Keep using ChatGPT or Claude for your own work. Your agents can run on the same models, and 100+ more. Switch with one line.
Six lines, or none at all.
As much or as little as you like. Here's the same support agent, with a knowledge base, memory and PII protection, built three ways:
agent.py
1from custodian_labs import Custodian23model = Custodian(4 model="gpt-4o",5 system_prompt="You are a helpful assistant...",6)78model.deploy()No. The Guardian Layer steps in first.
Every message is scanned for sensitive data (PII) before any model sees it. Names, emails, amounts and your own custom entities are masked, redacted or swapped for realistic stand-ins, then the originals are relinked in the response you get back.
The Guardian Layer is on by default. You can turn it off for any agent, but sensitive data then goes to the model as-is, at your own risk.
See how the Guardian Layer worksYou choose the degree of protection, without losing the context.
Most privacy tools only strip PII, and stripping context breaks AI accuracy. With the Guardian Layer you set the degree of protection: remove it completely, mask it, or transform it with our research-backed method that keeps the context and links the originals back. Nothing sensitive leaves your environment.
Research-backedSwaps PII for realistic stand-ins, so the model keeps the full context, then links the originals back into the reply.
Any of 100+, including Claude, GPT, Gemini, Mistral and open-source models.
Pick a model per agent, or bring your own API key. No code needed: just click the model you want in the builder. In code, it's a one-line change. Either way, your agent logic, knowledge base and guardrails stay exactly as they are.
Agent builder
Model
agent.py
1model="gpt-4o",2model="claude-fable-5",No. There's nothing to host.
Upload PDFs, Word docs, spreadsheets, CSV or text files and Custodian handles chunking, embeddings, retrieval and citations. Deploying gives you a hosted chat endpoint with session memory built in.
Yes. Switch tools on per agent.
Agents can search the web, read and send Gmail, post to X, scrape pages and parse documents. Anything else plugs in through MCP.
Python and JavaScript / TypeScript, or no code at all.
Both SDKs cover the same features: agents, knowledge bases, tools, teams and the Guardian Layer. The JavaScript SDK runs in Node, browsers and edge runtimes.
$ pip install custodian-labs$ npm install @custodianlabs/sdkYes. Start free, and upgrade when you're ready.
Simple, transparent pricing. Start today for free and scale when you're ready. There's no infrastructure to pay for on top.
- 100k tokens
- 1k requests
- GPT-4o
- Email and community support
Starter
Was $49now $19/ month
For users who want more limits, priority access, and VIP support
Get API Key- 1m tokens
- 10k requests
- All models + newest first
- Priority 24/7 support
- Direct roadmap input
- Monthly 1:1 product call
- Free compliance assessment
- Custom token limit
- Custom request limit
- On-premise or private cloud
- Offline deployments
- Custom compliance & data contracts
- Dedicated support & SLA
You've been missing out!
Our stack is already used by organisations ranging from Y Combinator companies to fast-growing AI-native startups. If you need custom limits, a private deployment or specific compliance terms, we'll shape Custodian around your stack.
Your first secure agent is 2 minutes away.
Describe it, code it, or build it visually, with the Guardian Layer on from day one. Start free, with no infrastructure to run.
Prefer code?