1  ---
  2  title: Self Hosting Overview
  3  description: Deploy and manage your own MLflow instance. Open-source, vendor-neutral MLOps and LLMOps platform for experiment tracking, model registry, and LLM observability.
  4  sidebar_position: 1
  5  ---
  6  
  7  # Self-Hosting MLflow
  8  
  9  > #### **_The most vendor-neutral MLOps/LLMOps platform in the world._**
 10  
 11  MLflow is fully open-source. Thousands of users and organizations run their own MLflow instances to meet their specific needs. Being open-source and trusted by the popular cloud providers, MLflow is the best choice for teams/organizations that worry about vendor lock-in.
 12  
 13  :::warning Default Storage Backend Change
 14  
 15  As of MLflow 3.7.0, the default tracking backend has changed from file-based storage (`./mlruns`) to SQLite database (`sqlite:///mlflow.db`) for better performance and reliability.
 16  
 17  **Existing users:** If you have existing data in `./mlruns`, MLflow will automatically detect and continue using it. No action is required.
 18  
 19  **New users:** New MLflow servers will use SQLite by default. To use file-based storage instead, set `MLFLOW_TRACKING_URI=./mlruns` or specify `--backend-store-uri ./mlruns` when starting the server.
 20  
 21  For more details and migration guidance, see [GitHub Issue #18534](https://github.com/mlflow/mlflow/issues/18534).
 22  
 23  :::
 24  
 25  ## The Quickest Path: Run `mlflow` Command
 26  
 27  The easiest way to start MLflow server is to run the `mlflow` CLI command in your terminal. This is suitable for personal use or small teams.
 28  
 29  First, install MLflow with:
 30  
 31  ```bash
 32  pip install mlflow
 33  ```
 34  
 35  :::info
 36  See [Secure Installs](/self-hosting/security/secure-installs) to learn how to pin dependencies to known good versions using hash checking and upload-time filtering.
 37  :::
 38  
 39  Then, start the server with:
 40  
 41  ```bash
 42  mlflow server --port 5000
 43  ```
 44  
 45  This will start the server and UI at `http://localhost:5000` using SQLite as the backend store (the default). You can connect the client to the server by setting the tracking URI:
 46  
 47  ```python
 48  import mlflow
 49  
 50  mlflow.set_tracking_uri("http://localhost:5000")
 51  
 52  # Start tracking!
 53  # Open http://localhost:5000 in your browser to view the UI.
 54  ```
 55  
 56  Now, you are ready to start your experiment!
 57  
 58  - [Tracing QuickStart](/genai/tracing/quickstart/)
 59  - [LLM Evaluation Quickstart](/genai/eval-monitor/quickstart/)
 60  - [Prompt Management Quickstart](/genai/prompt-registry/#getting-started)
 61  - [Model Training Quickstart](/ml/tracking/quickstart/)
 62  
 63  :::tip
 64  
 65  For production deployments or custom backend configurations, see <ins>[Backend Store](./architecture/backend-store)</ins> documentation.
 66  
 67  :::
 68  
 69  ## Other Deployment Options
 70  
 71  ### Docker Compose
 72  
 73  The MLflow repository includes a ready-to-run Compose project under `docker-compose/` that provisions MLflow, PostgreSQL, and [RustFS](https://github.com/rustfs/rustfs).
 74  
 75  ```bash
 76  git clone https://github.com/mlflow/mlflow.git
 77  cd mlflow/docker-compose
 78  cp .env.dev.example .env
 79  docker compose up -d
 80  # Open http://localhost:5000 in your browser to view the UI.
 81  ```
 82  
 83  Read the instructions [here](https://github.com/mlflow/mlflow/tree/master/docker-compose) for more details and configuration options for the docker compose bundle.
 84  
 85  ### Kubernetes
 86  
 87  To deploy on Kubernetes, use the MLflow Helm chart provided by [Bitnami](https://artifacthub.io/packages/helm/bitnami/mlflow) or [Community Helm Charts](https://artifacthub.io/packages/helm/community-charts/mlflow).
 88  
 89  ### Cloud Services
 90  
 91  If you are looking for production-scale deployments without maintenance costs, MLflow is also available as managed services from popular cloud providers.
 92  
 93  - [Databricks](https://www.databricks.com/product/managed-mlflow)
 94  - [AWS Sagemaker](https://aws.amazon.com/sagemaker/ai/experiments/)
 95  - [Azure Machine Learning](https://learn.microsoft.com/en-us/azure/machine-learning/concept-mlflow?view=azureml-api-2)
 96  - [Nebius](https://nebius.com/services/managed-mlflow)
 97  - [GCP (GKE)](https://gke-ai-labs.dev/docs/tutorials/frameworks-and-pipelines/mlflow/)
 98  
 99  ## Architecture
100  
101  MLflow, at a high level, consists of the following components:
102  
103  1. **Tracking Server**: The lightweight FastAPI server that serves the MLflow UI and API.
104  2. **Backend Store**: The Backend Store is relational database (or file system) that stores the metadata of the experiments, runs, traces, etc.
105  3. **Artifact Store**: The Artifact Store is responsible for storing the large artifacts such as model weights, images, etc.
106  
107  Each component is designed to be pluggable, so you can customize it to meet your needs. For example, you can start with a single host mode with SQLite backend and local file system for storing artifacts. To scale up, you can switch backend store to PostgreSQL cluster and point artifact store to cloud storage such as S3, GCS, or Azure Blob Storage.
108  
109  To learn more about the architecture and available backend options, see [Architecture](./architecture/overview).
110  
111  ## Workspaces
112  
113  MLflow supports [workspaces](/self-hosting/workspaces) to organize experiments, registered models, prompts, and artifacts on a shared MLflow instance. Workspaces add logical separation and workspace-level permissions so teams can collaborate without running separate servers. Workspaces are opt-in and require a SQL database backend.
114  
115  ## Access Control & Security
116  
117  MLflow support [username/password login](./security/basic-http-auth) via basic HTTP authentication, [SSO (Single Sign-On)](./security/sso), and [custom authentication plugins](./security/custom).
118  
119  MLflow also provides built-in [network protection](./security/network) middleware to protect your tracking server from network exposure.
120  
121  :::tip Try Managed MLflow
122  
123  Need highly secure MLflow server? Check out <ins>[Databricks Managed MLflow](https://www.databricks.com/product/managed-mlflow)</ins> to get fully managed MLflow servers with unified governance and security.
124  
125  :::
126  
127  ## FAQs
128  
129  See [Troubleshooting & FAQs](./troubleshooting) for more information.
130  
131  :::info[ACCESS DENIED?]
132  
133  When using the remote tracking server, you may hit an access denied error when accessing the MLflow UI
134  from a browser.
135  
136  > Invalid Host header - possible DNS rebinding attack detected
137  
138  This error typically indicates that the tracking server's network security settings need to be configured.
139  The most common causes are:
140  
141  - **Host validation**: The `--allowed-hosts` flag restricts which Host headers are accepted
142  - **CORS restrictions**: The `--cors-allowed-origins` flag controls which origins can make API requests
143  
144  To resolve this, configure your tracking server with the appropriate flags. For example:
145  
146  ```bash
147  mlflow server --allowed-hosts "mlflow.company.com,localhost:*" \
148                --cors-allowed-origins "https://app.company.com"
149  ```
150  
151  **Note**: These security options are only available with the default FastAPI-based server (uvicorn). They are
152  not supported when using Flask directly or with `--gunicorn-opts` or `--waitress-opts`.
153  
154  Refer to the <ins>[Network Security Guide](/self-hosting/security/network)</ins> for detailed configuration options.
155  
156  :::