Self-Host Langflow with Docker and Build AI Workflows

Updated 10 November 2026 · Standard formula, shown below

Self-hosting Langflow with Docker gives you a browser-based workspace for creating and running visual AI workflows on infrastructure you control. You’ll start the container, open Langflow locally, connect a model provider, and build a simple workflow; have Docker installed and an API key ready if your workflow will call a hosted model.

What You’ll Need

A computer or server with Docker Engine and Docker Compose available. Docker Desktop includes both on supported desktop systems.
A terminal and permission to run Docker commands.
A current Langflow Docker image. The commands below use the published langflowai/langflow image; check the project’s official documentation for the recommended tag and release notes before deploying.
An API key for a model provider if you plan to use a hosted language model. You can also configure a locally available model if you have an appropriate integration and resources.
For a persistent or shared deployment: a location for durable application data, a plan for backups, and a secure way to manage credentials.

Step-by-Step

Check Docker is working. Open a terminal and run docker --version and docker compose version. If a command is not found, install or update Docker, start its service or desktop app, and reopen the terminal. On Linux, verify your account has permission to use Docker or use the appropriate administrator privileges.
Choose how the service will be reached. For a private test on the same computer, bind Langflow to the loopback address so it is not exposed to your network. For a server deployment, do not simply publish the interface to the public internet. Put it behind a properly configured HTTPS reverse proxy, restrict access, and follow Langflow’s deployment security guidance.
Make a project directory. Create a folder for the deployment, then open a terminal in it. For example, run mkdir langflow-local followed by cd langflow-local. Keeping the compose file and configuration together makes updates and recovery easier.
Create a Compose file. Save the following as compose.yaml for a local-only test:
```
services:
  langflow:
    image: langflowai/langflow:latest
    ports:
      - "127.0.0.1:7860:7860"
    volumes:
      - langflow-data:/app/langflow
    restart: unless-stopped
volumes:
  langflow-data:
```
This maps the container’s web port to port 7860 on your computer and stores data in a named Docker volume. The image tag latest is convenient for a trial, but can change over time; for a stable deployment, select a specific supported release tag after checking the project documentation. Confirm the documented data directory for the release you choose before relying on the volume for persistence.
Start Langflow. From the directory containing compose.yaml, run docker compose up -d. Docker downloads the image if needed and starts the container in the background. Check its status with docker compose ps; use docker compose logs -f langflow to watch startup messages. Wait for the application to finish starting before opening the browser.
Open the interface. Visit http://localhost:7860 on the same computer. If you changed the host port, use that port instead. Follow any first-run prompts shown by your installed version. Keep the local-only port binding for personal testing; it does not provide public HTTPS or user access controls.
Create a basic flow. Choose the option to create a flow or open a blank canvas. Add a chat input or prompt component, a language-model component, and a chat output component. Connect the output of each component to the compatible input of the next, following the ports shown in the interface. Component names and available integrations can vary by Langflow release.
Configure the model safely. Select a provider and model you can access, then add its API key through Langflow’s supported credential or environment-variable mechanism. Do not paste secrets into a flow that will be shared or commit them to a Compose file or source-control repository. If the interface offers a credential store, use it and restrict who can view or edit the flow.
Run and verify the workflow. Enter a short test prompt and run the flow. Check that each component receives the expected input and that the final output is sensible. If it fails, inspect the component error and container logs; verify the key, provider access, model selection, and connections before changing unrelated settings.
Save, stop, and update deliberately. Save the flow in the interface and confirm it remains after restarting the container. Stop the service with docker compose down; the named volume remains unless you explicitly remove it. Before upgrading, back up important data, check release notes for migration or configuration changes, update the image tag, and run docker compose up -d again. Avoid docker compose down -v unless you intend to delete the named volume and its data.

Common Mistakes

Using the wrong address or port: With the example file, the browser address is http://localhost:7860. A server’s private or public address is not interchangeable with localhost.
Publishing the interface without protection: Port mapping can make a service reachable beyond your computer. Keep the loopback binding for local use; for remote access, configure authentication, network restrictions, and HTTPS rather than exposing an unprotected interface.
Assuming container storage is permanent: Data written only inside a container can disappear when it is replaced. Use the documented persistent storage location and verify persistence by restarting before adding important workflows.
Putting API keys in shared files: Compose files can be copied, backed up, or committed. Keep credentials out of repositories and shared flow exports, and rotate a key if it is exposed.
Expecting every component to work offline: Hosted model components need network access and valid provider credentials. A running Langflow container alone does not include access to a hosted model.
Updating blindly: A moving image tag may pull a release with changes. Pin a release for repeatable deployments and review upgrade instructions before replacing a working installation.

Tips

Start with a small flow and test it after each connection. This makes configuration errors easier to locate.
Name flows and components clearly, and add notes for required model settings or environment variables.
Keep a copy of your Compose file and make a separate, verified backup of persistent data. A volume is storage, not a backup.
Use a non-production test key while experimenting, and avoid sending sensitive data to a model unless your provider and configuration are appropriate for it.
For a shared server, follow Langflow’s current deployment documentation for authentication, secrets, persistent storage, reverse proxy settings, and upgrades. Do not assume the local test configuration is production-ready.

Troubleshooting

The page will not load: Run docker compose ps and check whether the service is running. Read docker compose logs --tail=100 langflow for startup errors. Confirm port 7860 is not already in use; if it is, change the host side of the port mapping, for example to 127.0.0.1:7861:7860, then visit http://localhost:7861.

The container exits or keeps restarting: Check the logs for configuration, permission, or resource errors. Confirm the image tag is valid and the computer has enough available memory. Try a documented supported image tag rather than guessing at environment variables or mount paths.

A flow reports a provider or model error: Check that the credential is present, correctly entered, and permitted to use the selected model. Confirm the container can reach the provider and that any account-side access requirements are met. Do not include the full secret when sharing logs.

Flows or settings disappear after an update: Verify that the named volume is still declared and mounted at the data path required by your Langflow release. Check that you did not remove the volume with docker compose down -v. Restore from a known-good backup if necessary, and consult the release-specific migration guidance.

Another device cannot connect: The sample configuration intentionally listens only on your own computer. For remote access, use a secured deployment design and inspect firewall and proxy settings. Do not change the binding to all network interfaces unless you have put appropriate access controls in place.

Frequently Asked Questions

Is Langflow free to self-host?
Docker can run the published Langflow image without a managed Langflow service, but model providers, server resources, and other integrations may have separate costs or requirements. Check the current project license and provider terms.

Can I run Langflow without Docker?
Other installation methods may be available, but this guide uses Docker to isolate the application and make startup and updates easier to manage.

Does Langflow include a language model?
Do not assume it does. A workflow generally needs a configured provider or model integration, and hosted models require valid credentials and network access.

Will my flows survive a restart?
They should when the release’s application data is stored in a correctly mounted persistent volume. Verify the data path for your chosen image and test a restart before relying on it.

Can I access my self-hosted instance from another computer?
Not with the loopback-only example configuration. Remote access requires deliberate network configuration and suitable authentication and HTTPS protections.

How do I update Langflow?
Back up persistent data, review the release notes and migration instructions, select the intended image tag, then recreate the service with Docker Compose. Keep the previous backup until the upgraded instance and workflows are verified.

Method: the standard formula for this calculation, rounded as each result states. See how the calculators work or report an error.