C4C Private: lives only in your environment. No outside connections.
The same regulation library behind the C4C Connector, running inside your own network or AWS account. Your AI tools and agents read the regulator’s exact words from your own server. Nothing about your work reaches us, and rule updates arrive through your own gates.
A library, not an agent
Most AI tools raise security questions because they act: they call out, write data, or read your files. C4C Private does none of that. At its core it is the regulators’ text, formatted so AI can read it efficiently, with a small lookup server in front. In your environment it:
- Only reads. It cannot write to your systems, take actions or change anything.
- Never sees your documents. Your AI asks it for a topic or a citation; your files stay in your AI tool.
- Has no outside connections. Questions, answers and usage never leave your network, and the server never needs to contact us.
- Cannot be quietly altered. Every release is signed. The server checks the signature before installing anything and refuses what does not verify.
One honest boundary: the risk that remains sits in the AI tool you connect, not in C4C. A cloud assistant (ChatGPT, a Claude connector, Copilot Studio) reaches your server from its vendor’s cloud. For nothing to leave your network, use an in-network agent or model, or add C4C Guardrails.
Machine-readable, so your AI can actually read the rule
Most agency guides are too big for an AI to read whole. In our measurement of all 11,134 current sections, 43% of mortgage rules sit in documents larger than an AI’s context window. C4C hands your AI the one section it needs, about 200 tokens on average, with its citation, version and effective date.
| Source document | Approximate size | One C4C section (median) |
|---|---|---|
| Freddie Mac Seller/Servicer Guide | 1.9 million tokens | about 860 tokens |
| FHA Handbook 4000.1 | 1 million tokens | about 330 tokens |
| Fannie Mae Selling Guide | 650,000 tokens | about 650 tokens |
| Fannie Mae Servicing Guide | 410,000 tokens | about 200 tokens |
Token counts are estimates (characters divided by four), measured against the official downloadable documents. How we measured →
Updates through your own gates
You choose how rule updates arrive. Each release is signed, so it is verified however it travels.
- Your own mirror (no contact with us). Your IT fetches each release at your gate, scans or approves it, and serves it from an internal server or an S3 bucket in your AWS account. Your C4C Private servers update from there, with no license key and no connection to us.
- Direct. The server fetches signed releases from us over HTTPS, sending only a free license key.
- Offline. Load the signed release files by hand.
If updates stop, the server keeps answering and marks the library stale after 14 days, so nobody mistakes an old rule for a current one.
Get C4C Private
- Docker, anywhere. Build from the public source and install guide on GitHub. Needs Docker, 2 GB of memory and about 5 GB of disk.
- AWS Marketplace. A free container listing for Amazon ECS and EKS, including the S3 mirror setup, is in review with AWS.
- Free license key. Sign up with a work email; it is used only to fetch releases, at your gate or on the server.
The code is public under the Elastic License 2.0. The regulatory library is free for research and reference use under the C4C Library License. Research only, not legal advice.
C4C Sandbox: C4C Private with guardrails, set up and supported
The C4C Sandbox is your own environment where AI does compliance work with mortgage guardrails on:
- C4C Private in your network or AWS account, updated through your gates
- C4C Guardrails: decision records while staff work, and the C4C Monitor checking your AI records after the fact
- C4C Playbooks, including Fannie Mae AI governance (Supplement Section 4)
- Setup, support and an update commitment