> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tracebloc.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Overview

> How tracebloc runs on your infrastructure — and why your data never leaves it.

**Models come to your data — your data never leaves your infrastructure.** You run a tracebloc *secure environment* on your own hardware. Peers submit models that are scanned and run in isolation against your data, on your machines. What leaves is results, model weights, dataset metadata and [usage events](https://docs.tracebloc.io/environment-setup/security.md#usage-events): after each training cycle your secure environment sends the weight updates to tracebloc, which combines them across secure environments.

## The trust boundary

![How tracebloc keeps your data on your infrastructure — models come to the data; results and encrypted weight updates leave](https://mintcdn.com/tracebloc/EVB0IGskWPJcKhjR/images/traceblocfederatedlearning.png?fit=max\&auto=format\&n=EVB0IGskWPJcKhjR\&q=85\&s=25bcf5a575b409f8d337be15cefde5cc)

Raw data stays inside your infrastructure. Your secure environment opens an **outbound-only** connection — nothing reaches in to pull your data out.

## How it works

**1. Deploy your secure environment**

One command sets up your secure environment on your own infrastructure — a laptop, an on-prem server, or a cloud cluster.

**2. Ingest your datasets**

Stage your training and test data locally. Metadata syncs to the platform so peers can see what's available — **the raw data never moves.**

**3. Create a use case**

Define a use case from your datasets and set how submitted models are evaluated.

**4. Invite peers**

Whitelist peers by email. Only the people you invite can take part.

**5. Peers submit and train models**

Each model is **scanned for vulnerabilities**, then trains against your data in an **isolated container**, on your hardware.

**6. Results and weight updates are shared**

Training and evaluation results flow back to you over TLS. The model's weight updates go to tracebloc after each training cycle, where they are combined. Whether data scientists can download trained weights is your choice: the **Weight Files** setting in the use case's admin panel (on by default).

## What stays, what leaves

| Data | Shared with the platform? |
| - | - |
| Raw training & test data | **Never** — it stays on your infrastructure |
| Dataset metadata: title, schema, row and per-class counts, and a preview of up to 10 record IDs with their labels | Yes — so peers know what's available. No file contents or feature values |
| Training & evaluation results | Yes — the metrics models are judged on |
| Trained model weights | **Yes, to tracebloc** after each cycle, for averaging. Downloads by data scientists follow the use case's **Weight Files** setting (on by default) |

Usage events from tracebloc's services, the installer and the CLI also leave: a name and its declared fields, never the rest of the log line. Every other log line stays in your secure environment. An install setting stops them — see [Security & data handling](https://docs.tracebloc.io/environment-setup/security.md#usage-events).

Enforced by per-job container isolation, a NetworkPolicy that restricts training-pod egress, a vulnerability scan before any model runs, and TLS on all traffic.

> [!NOTE]
>
> **The mental model:** `1 machine = 1 secure environment = n datasets`. One deployment per machine; as many datasets inside it as you like.

## Vocabulary

| Term | What it is |
| - | - |
| **Secure environment** | Your private, dedicated AI environment — tracebloc's software running on your own infrastructure, where you invite peers to train models on your data |
| **Client ID** | The credential that connects your secure environment to the platform (the installer creates it when it registers your machine) |
| **Dataset** | Data you've staged locally for use cases |
| **Use case** | A task peers build models for, against your datasets |
| **Peer** | An external data scientist who submits models — and never sees your raw data |

## What it touches

On Linux the installer adds native k3s, with no Docker; what it needs and adds there is in [Linux prerequisites](https://docs.tracebloc.io/environment-setup/deploy-local.md#linux). On a Mac with Apple silicon and macOS 26 or later it adds k3s in Apple's `container` runtime, with no Docker, no Homebrew and no administrator rights; what it needs and adds there is in [Mac prerequisites](https://docs.tracebloc.io/environment-setup/deploy-local.md#macos). On Windows it adds k3s in tracebloc's own WSL distro, with no Docker Desktop; what it needs and adds there is in [Windows prerequisites](https://docs.tracebloc.io/environment-setup/deploy-local.md#windows). Everywhere it adds the `tracebloc` CLI and `~/.tracebloc`, and switches kubectl's current context to the new cluster. On Windows that context is in your Windows kubeconfig, and `kubectl` itself is inside the distro: see [kubectl and Helm on Windows](https://docs.tracebloc.io/environment-setup/operations.md#kubectl-and-helm-on-windows). The full list is under [Known footprint](https://docs.tracebloc.io/environment-setup/security.md#how-it-is-enforced). Uninstalling is a single command, `tracebloc delete`, and your data is yours throughout.

## Get started

> [card] [Quick Start](https://docs.tracebloc.io/environment-setup/quickstart.md)
>
> A running secure environment in about 10 minutes, with one command.

> [card] [Deployment environments](https://docs.tracebloc.io/environment-setup/deployment-environments.md)
>
> Deploy on a local machine, bare-metal, EKS, AKS, or OpenShift.

> [!NOTE]
>
> Want the guarantees in detail for your security team? See [Security & data handling](https://docs.tracebloc.io/environment-setup/security.md).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.