To run a Bokeh application on a Bokeh server from a single Python script, pass the script name to bokeh serve on the command line:

bokeh serve

By default, the Bokeh application will be served by the Bokeh server on a default port (5006) at localhost, under the path /app_script, i.e.,


It is also possible to run the same commmand with jupyter notebooks:

bokeh serve app_notebook.ipynb

This will generate the same results as described with a python script and the application will be served on a default port (5006) at localhost, under the path /app_notebook

Applications can also be created from directories. The directory should contain a (and any other helper modules that are required) as well as any additional assets (e.g., theme files). Pass the directory name to bokeh serve to run the application:

bokeh serve app_dir

It is possible to run multiple applications at once:

bokeh serve app_dir

If you would like to automatically open a browser to display the HTML page(s), you can pass the --show option on the command line:

bokeh serve app_dir --show

This will open two pages, for /app_script and /app_dir, respectively.

If you would like to pass command line arguments to Bokeh applications, you can pass the --args option as the LAST option on the command line:

bokeh serve --args foo bar --baz

Everything that follows --args will be included in sys.argv when the application runs. In this case, when executes, the contents of sys.argv will be ['', 'foo', 'bar', '--baz'], consistent with standard Python expectations for sys.argv.

Note that if multiple scripts or directories are provided, they all receive the same set of command line arguments (if any) given by --args.

If you have only one application, the server root will redirect to it. Otherwise, You can see an index of all running applications at the server root:


This index can be disabled with the --disable-index option, and the redirect behavior can be disabled with the --disable-index-redirect option.

Another way to run multiple applications is using glob notation to indicate that all the files matching a particular pattern should be served.

bokeh serve *.py

Command line shells will normally expand the *.py automatically. However, if you are starting a Bokeh server programmatically, then filename arguments with globs may not be expanded by the shell. In situations like this, the --glob flag may be used to make the Bokeh server explicitly perform globbing:["bokeh", "serve", "--glob", "*.py"])

A .ico file may be configured with the --ico-path option, and the Bokeh server will return the contents of this file for /favicon.ico:

bokeh serve --ico-path=/assets/app.ico

Since browsers only look at the top-level root, only one .ico file may be supplied, even if multiple apps are running. By default, a Bokeh project logo is returned as the favicon.ico file.

Application Configuration#

Bokeh server can fork the underlying tornado server into multiprocess. This is useful when trying to handle multiple connections especially in the context of apps which require high computational loads. Default behavior is one process. using 0 will auto-detect the number of cores and spin up corresponding number of processes

bokeh serve --num-procs 2

Note that due to limitations inherent in Tornado, Windows does not support --num-procs values greater than one! In this case consider running multiple Bokeh server instances behind a load balancer.

The Bokeh server can also add an optional prefix to all URL paths. This can often be useful in conjunction with “reverse proxy” setups.

bokeh serve --prefix foobar

Then the application will be served under the following URL:


If needed, Bokeh server can send keep-alive pings at a fixed interval. To configure this feature, set the --keep-alive option:

bokeh serve --keep-alive 10000

The value is specified in milliseconds. The default keep-alive interval is 37 seconds. Give a value of 0 to disable keep-alive pings.

Network Configuration#

To control the port that the Bokeh server listens on, use the --port argument:

bokeh serve --port 8080

To listen on an arbitrary port, pass 0 as the port number. The actual port number will be logged at startup.

Similarly, a specific network address can be specified with the --address argument. For example:

bokeh serve --address

will have the Bokeh server listen all available network addresses.

By default, cross site connections to the Bokeh server websocket are not allowed. You can enable websocket connections originating from additional hosts by specifying them with the BOKEH_ALLOW_WS_ORIGIN environment variable or the --allow-websocket-origin option:

bokeh serve --allow-websocket-origin

It is possible to specify multiple allowed websocket origins by adding the --allow-websocket-origin option multiple times and to provide a comma separated list of hosts to BOKEH_ALLOW_WS_ORIGIN

To have the Bokeh server override the remote IP and URI scheme/protocol for all requests with X-Real-Ip, X-Forwarded-For, X-Scheme, X-Forwarded-Proto headers (if they are provided), set the --use-xheaders option:

bokeh serve --use-xheaders

This is typically needed when running a Bokeh server behind a reverse proxy that is SSL-terminated.


It is not advised to set this option on a Bokeh server directly facing the Internet.

A Bokeh server can also terminate SSL connections directly by specifying the path to a single file in PEM format containing the certificate as well as any number of CA certificates needed to establish the certificate’s authenticity:

bokeh serve --ssl-certfile /path/to/cert.pem

Alternatively, the path may also be supplied by setting the environment variable BOKEH_SSL_CERTFILE.

If the private key is stored separately, its location may be supplied by setting the --ssl-keyfile command line argument, or by setting the BOKEH_SSL_KEYFILE environment variable. If a password is required for the private key, it should be supplied by setting the BOKEH_SSL_PASSWORD environment variable.

Session ID Options#

Typically, each browser tab connected to a Bokeh server will have its own session ID. When the server generates an ID, it will make it cryptographically unguessable. This keeps users from accessing one another’s sessions.

To control who can use a Bokeh application, the server can sign session IDs with a secret key and reject “made up” session names. There are three modes, controlled by the --session-ids argument:

bokeh serve --session-ids signed

The available modes are: unsigned, signed or external-signed

In unsigned mode, the server will accept any session ID provided to it in the URL. For example, http://localhost/app_script?bokeh-session-id=foo will create a session foo. In unsigned mode, if the session ID isn’t provided with ?bokeh-session-id= in the URL, the server will still generate a cryptographically-unguessable ID. However, the server allows clients to create guessable or deliberately-shared sessions if they want to.

unsigned mode is most useful when the server is running locally for development, for example you can have multiple processes access a fixed session name such as default. unsigned mode is also convenient because there’s no need to generate or configure a secret key.

In signed mode, the session ID must be in a special format and signed with a secret key. Attempts to use the application with an invalid session ID will fail, but if no ?bokeh-session-id= parameter is provided, the server will generate a fresh, signed session ID. The result of signed mode is that only secure session IDs are allowed but anyone can connect to the server.

In external-signed mode, the session ID must be signed but the server itself won’t generate a session ID; the ?bokeh-session-id= parameter will be required. To use this mode, an external process (such as another web app) would use the function bokeh.util.token.generate_session_id() to create valid session IDs. The external process and the Bokeh server must share the same BOKEH_SECRET_KEY environment variable.

external-signed mode is useful if you want another process to authenticate access to the Bokeh server. If someone is permitted to use a Bokeh application, you would generate a session ID for them, then redirect them to the Bokeh server with that valid session ID. If you don’t generate a session ID for someone, then they can’t load the app from the Bokeh server.

In both signed and external-signed mode, the secret key must be kept secret; anyone with the key can generate a valid session ID.

The secret key should be set in a BOKEH_SECRET_KEY environment variable and should be a cryptographically random string with at least 256 bits (32 bytes) of entropy. The bokeh secret command can generate new secret keys.

Authentication Options#

The Bokeh server can be configured to only allow connections in case there is a properly authenticated user. This is accomplished by providing the path to a module that implements the necessary functions on the command line:

bokeh serve --auth-module=/path/to/

or by setting the BOKEH_AUTH_MODULE environment variable.

The module must contain one of the following two functions that will return the current user (or None):

def get_user(request_handler):

async def get_user_async(request_handler):

The function is passed the Tornado RequestHandler and can inspect cookies or request headers to determine the authenticated user. If there is no valid authenticated user, these functions should return None.

Additionally, the module must specify where to redirect unauthenticated users. It must contain either:

  • a module attribute login_url and (optionally) a LoginHandler class

  • a function definition for get_login_url

login_url = "..."

class LoginHandler(RequestHandler):

def get_login_url(request_handler):

When a relative login_url is given, an optional LoginHandler class may also be provided, and it will be installed as a route on the Bokeh server automatically.

The get_login_url function is useful in cases where the login URL must vary based on the request, or cookies, etc. It is not possible to specify a LoginHandler when get_url_function is defined.

Analogous to the login options, optional logout_url and LogoutHandler values may be define an endpoint for logging users out.

If no auth module is provided, then a default user will be assumed, and no authentication will be required to access Bokeh server endpoints.


The contents of the auth module will be executed!

Bokeh can also enable the use of Tornado’s XFRF cookie protection. To turn this feature on, use the --enable-xsrf-cookies option, or set the environment variable BOKEH_XSRF_COOKIES=yes. If this setting is enabled, any PUT, POST, or DELETE operations on custom or login handlers must be instrumented properly in order to function. Typically, this means adding the xsrf_form_html() module to HTML form submissions templates. For full details, see:

Session Expiration Options#

To configure how often to check for unused sessions. set the --check-unused-sessions option:

bokeh serve --check-unused-sessions 10000

The value is specified in milliseconds. The default interval for checking for unused sessions is 17 seconds. Only positive integer values are accepted.

To configure how often unused sessions last. set the --unused-session-lifetime option:

bokeh serve --unused-session-lifetime 60000

The value is specified in milliseconds. The default lifetime interval for unused sessions is 15 seconds. Only positive integer values are accepted.

Diagnostic Options#

The logging level can be controlled by the --log-level argument:

bokeh serve --log-level debug

The available log levels are: trace, debug, info, warning, error or critical

The log format can be controlled by the --log-format argument:

bokeh serve --log-format "%(levelname)s: %(message)s"

The default log format is "%(asctime)s %(message)s"

To control how often statistic logs are written, set the --stats-log-frequency option:

bokeh serve --stats-log-frequency 30000

The value is specified in milliseconds. The default interval for logging stats is 15 seconds. Only positive integer values are accepted.

Bokeh can also optionally log process memory usage. This feature requires the optional psutil package to be installed. To enable memory logging set the --mem-log-frequency option:

bokeh serve --mem-log-frequency 30000

The value is specified in milliseconds. The default interval for logging stats is 0 (disabled). Only positive integer values are accepted.

class Serve(parser: ArgumentParser)[source]#

Subcommand to launch the Bokeh server.

customize_applications(args: Namespace, applications: dict[str, Any]) dict[str, Any][source]#

Allows subclasses to customize applications.

Should modify and return a copy of the applications dictionary.

customize_kwargs(args: Namespace, server_kwargs: dict[str, Any]) dict[str, Any][source]#

Allows subclasses to customize server_kwargs.

Should modify and return a copy of the server_kwargs dictionary.

customize_server(server: Server) Server[source]#

Allows subclasses to customize the server.

Should apply modifications to the server and wrap it or return the same instance.

invoke(args: Namespace) None[source]#
name: ClassVar[str] = 'serve'#

name for this subcommand