Reference
Client SDK
Python

Python Client API

💡

Before going through this guide, make sure you follow the Oso Cloud Quickstart to get your Oso Cloud API Key properly set in your environment.

First, install the oso-cloud package on PyPI:


pip install oso-cloud

Instantiating an Oso Cloud client

The Oso Cloud client provides an Oso class that takes your Oso Cloud URL and API key:


from oso_cloud import Oso
oso = Oso(url="https://cloud.osohq.com", api_key=YOUR_API_KEY)
# Later:
oso.insert(("has_role", user, role, resource))
# Wherever authorization needs to be performed:
if oso.authorize(user, action, resource):
# Action is allowed

ℹ️

You should instantiate one client and share it across your application. Under the hood, it reuses connections to avoid paying the cost of negotiating a new connection on every request.

Specifying an Oso Fallback host

If you have deployed Oso Fallback nodes to your infrastructure, you may specify the host when instantiating the Oso Cloud client.


# Assumes Oso Fallback is hosted at http://localhost:8080
oso = Oso(url="https://cloud.osohq.com", api_key=YOUR_API_KEY, fallback_url="http://localhost:8080")

Passing application entities into the client

Under the hood, Oso Cloud represents an entity in your application as a combination of a type and an ID, which together uniquely identify the entity. The Python client represents these entities with the Value class, which has both type and id fields. For example:


from oso_cloud import Value
alice = Value(type="User", id="alice")
anvils_repository = Value("Repository", "anvils") # shorthand

You will pass objects like these into nearly every function call you make to the Python client.

Primitive value conversion

Additionally, for fields with primitive values you may pass an instance of the corresponding Python type directly instead of a full Value object:


oso.insert(("is_weird", 10, "yes", True))
# equivalent to:
# (
# "is_weird",
# Value("Integer", "10"),
# Value("String", "yes"),
# Value("Boolean", "true")
# )

Management API

Add fact: oso.insert((name, *args))

Adds a fact named name with the provided arguments. Example:


oso.insert(("has_role", Value("User", "bob"), "owner", Value("Organization", "acme")))

Delete fact: oso.delete((name, *args))

Deletes a fact. Does not throw an error if the fact is not found. Example:


oso.delete(("has_role", Value("User", "bob"), "maintainer", Value("Repository", "anvils")))

When deleting facts, you can use None as a wildcard to delete many facts at once.


# remove all of bob's roles across all resources
oso.delete(("has_role", Value("User", "bob"), None, None))

You can also use oso_cloud.ValueOfType to delete facts with an argument of a particular type.


# remove all of bob's roles across all Repositories
oso.delete(("has_role", Value("User", "bob"), None, ValueOfType("Repository")))

Transactionally delete and add facts: oso.batch()

ℹ️

For Oso Cloud developer accounts, batch calls are limited to 20 facts. If you attempt to send more than 20 facts, these functions will throw an error.

Allows deleting and inserting many facts in one atomic transaction. Deletions and insertions are run in the order they appear in the context. Example:


bob = Value("User", "bob")
with oso.batch() as tx:
tx.insert(("has_role", bob, "owner", Value("Organization", "acme")))
tx.delete(("has_role", bob, "maintainer", Value("Repository", "anvils")))

As in calls to delete, you may use wildcards (None) and type-constrained wildcards (ValueOfType) in tx.delete calls.

List facts: oso.get((name, *args))

ℹ️

For Oso Cloud developer accounts, Get calls are limited to 1000 results. If you have more than 1000 facts, the function will throw an error.

Lists facts that are stored in Oso Cloud. Can be used to check the existence of a particular fact, or used to fetch all facts that have a particular argument:


bob = Value("User", "bob")
anvils = Value("Repository", "anvils")
# Get one fact:
oso.get(("has_role", bob, "admin", anvils))
# => [(
# "has_role",
# Value("User", "bob"),
# Value("String", "admin"),
# Value("Repository", "anvils")
# )]
# List all role-related facts on the `anvils` repo
oso.get(("has_role", None, None, anvils))
# => [(
# "has_role",
# Value("User", "bob"),
# Value("String", "admin"),
# Value("Repository", "anvils")
# ),
# ... other has_role facts
# ]

Note that None behaves like a wildcard: passing the fact arguments None, None, anvils means "find all facts where anvils is the third argument, regardless of other arguments".

You can also use oso_cloud.ValueOfType to get all facts with an argument of a certain type:


# List all role-related facts on any repo:
oso.get(("has_role", None, None, ValueOfType("Repository")))
# => [(
# "has_role",
# Value("User", "bob"),
# Value("String", "reader"),
# Value("Repository", "acme")
# ),
# ... other has_role facts
# ]

oso.get() only returns facts you've explicitly added. If you want to return a list of authorized resources, use the Check API. For example, to answer "on which resources can a given user perform a given action", use oso.list(). If you want to query for arbitrary information that can be derived from your facts and policy, use the Query Builder API.

Check API

ℹ️

For Oso Cloud developer accounts, the number of context facts per request is limited to 20; and the number of records returned is limited to 1000.

Context facts

You may provide an array of context facts as an optional argument to any of the Check API methods. When Oso Cloud performs a check, it will consider these context facts in addition to any other facts you've previously added. Context facts are only used in the API call in which they're provided— they do not persist across requests. Learn more about context facts.

Check a permission: oso.authorize(actor, action, resource)

Determines whether or not an action is allowed, based on a combination of authorization data and policy logic. Example:


alice = Value("User", "alice")
anvils_repository = Value("Repository", "anvils")
if not oso.authorize(alice, "read", anvils_repository):
raise Exception("Action is not allowed")

You may provide a list of context facts as an optional fourth argument to this method. Example:


issue_on_anvils_repository = Value("Issue", "anvils-1")
oso.authorize(alice, "read", anvils_repository, [
# a context fact
("has_relation", issue_on_anvils_repository, "parent", anvils_repository)
])

List authorized resources: oso.list(actor, action, resource_type)

Fetches a list of resources on which an actor can perform a particular action. Example:


alice = Value("User", "alice")
oso.list(alice, "read", "Repository")
# => ["acme"]

You may provide a list of context facts as an optional fourth argument to this method. Example:


anvils_repository = Value("Repository", "anvils")
acme_repository = Value("Repository", "acme")
issue_on_acme_repository = Value("Issue", "acme-1")
issue_on_anvils_repository = Value("Issue", "anvils-2")
oso.list(
alice, "read", "Issue",
[ # context facts
("has_relation", issue_on_anvils_repository, "parent", anvils_repository),
("has_relation", issue_on_acme_repository, "parent", acme_repository),
]
)
# => ["acme-1"]

List authorized actions: oso.actions(actor, resource)

Fetches a list of actions which an actor can perform on a particular resource. Example:


alice = Value("User", "alice")
acme_repository = Value("Repository", "acme")
oso.actions(alice, acme_repository)
# => ["read"]

You may provide a list of context facts as an optional third argument to this method. Example:


issue_on_acme_repository = Value("Issue", "acme-1")
actions = oso.actions(
alice, issue_on_acme_repository,
# a context fact
[
("has_relation", issue_on_acme_repository, "parent", acme_repository)
]
)
# => ["read"]

Query for any rule: oso.build_query((predicate, *args))

Query Oso Cloud for any predicate and any combination of concrete and wildcard arguments. Unlike oso.get, which only lists facts you've added, you can use oso.build_query to list derived information about any rule in your policy. Example:


actor = Value("User", "bob")
repository = typed_var("Repository");
# Query for all the repos `User:bob` can `read`
oso.build_query(("allow", actor, "read", repository)).evaluate(repository)
# => [ "acme", "anvils" ]

Query Builder API

The oso.build_query() API is a builder-style API where you chain methods to construct a query and then execute it.

oso.build_query((predicate, *args))

The oso.build_query() function takes the name of the rule you want to query and a list of arguments. The arguments can be concrete values (e.g., "read" or Value("User", "bob")) or type-constrained variables constructed via the typed_var function:


actor = Value("User", "bob")
repository = typed_var("Repository")
# Query for all the repositories bob can read
oso.build_query(("allow", actor, "read", repository))
# => QueryBuilder { ... }

Note: once you've finished building up your query, you must call evaluate to run it and get the results.

QueryBuilder.and_((predicate, *args))

This function adds another condition that must be true of the query results.

For example:


actor = Value("User", "bob")
repository = typed_var("Repository")
folder = Value("Folder", "folder-1")
# Query for all the repositories this user can read...
(
oso
.build_query(("allow", actor, "read", repository))
#... and require the repositories to belong to the given folder.
.and_(("has_relation", repository, "folder", folder))
)
# => QueryBuilder { ... }

Note: once you've finished building up your query, you must call evaluate to run it and get the results.

QueryBuilder.in_(variable, values)

This function requires a given typed_var query variable to be included in a given set of values. You can only call in_ once per variable per query. Calling in_ a second time with the same variable on the same query builder will throw an error.

For example:


actor = Value("User", "bob")
repositories = ["acme", "anvil"]
action = typed_var("String")
repository = typed_var("Repository")
# Query for all the actions this user can perform on any repository...
(
oso
.build_query(("allow", actor, action, repository))
# ...given that the repository's ID is in the given list of IDs.
.in_(repository, repositories)
)
# => QueryBuilder { ... }

Note: once you've finished building up your query, you must call evaluate to run it and get the results.

QueryBuilder.with_context_facts(context_facts)

This function adds the given context facts to the query. For example:


actor = Value("User", "bob")
repository = typed_var("Repository")
# Query for all the repositories bob can read...
(
oso
.build_query(("allow", actor, "read", repository))
# ...while including the fact that bob owns acme
.with_context_facts([
("has_role", actor, "owner", Value("Repository", "acme")),
])
)
# => QueryBuilder { ... }

For more information on context facts, see this section.

Note: once you've finished building up your query, you must call evaluate to run it and get the results.

QueryBuilder.evaluate()

This function evaluates the built query, fetching the results from Oso.

The return type of this function varies based on the arguments you pass in.

  • If you pass no arguments, this function returns a boolean. For example:


    allowed = (
    oso
    .build_query(("allow", actor, action, resource))
    .evaluate()
    )
    # => true if the given actor can perform the given action on the given resource

  • If you pass a single typed_var query variable, this function returns a list of values for that variable. For example:


    action = typed_var("String");
    actions = (
    oso
    .build_query(("allow", actor, action, resource))
    .evaluate(action)
    )
    # => all the actions the actor can perform on the given resource- eg. ["read", "write"]

  • If you pass a tuple of typed_var query variables, this function returns a list of tuples of values for those variables. For example:


    action = typed_var("String")
    repository = typed_var("Repository")
    pairs = (
    oso
    .build_query("allow", actor, action, repository)
    .evaluate((action, repository))
    )
    # => an array of pairs of allowed actions and repo IDs-
    # eg. [("read", "acme"), ("read", "anvil"), ("write", "anvil")]

  • If you pass a dict mapping one typed_var query variable (call it K) to another (call it V), returns a dict grouping unique values of K to unique values of V for each value of K. For example:


    action = typed_var("String")
    repository = typed_var("Repository")
    repo_actions = (
    oso
    .build_query(("allow", actor, action, repository))
    .evaluate({repository: action})
    )
    # => a dict of repo IDs to allowed actions-
    # eg. { "acme": ["read"], "anvil": ["read", "write"]}

Some queries have unconstrained results. For instance, maybe users with the admin role can read all Repository entities in your application. In this case, rather than returning an array containing the ID of every repository, evaluate will return an array containing the string "*". For example:


repos = typed_var("Repository")
(
oso
.build_query(("allow", Value("User", "admin"), "read", repos))
.evaluate(repos); # Return just the IDs of the repos admin can read
)
# => ["*"] # admin can read anything

Query Builder Examples

Field-level access control

actor = Value("User", "alice")
resource = Value("Repository", "anvil")
field = typed_var("Field")
results = (
oso
.build_query(("allow_field", actor, "read", resource, field))
.evaluate(field)
)
# => Returns a list of the fields alice can read on the given repo- eg.
# ["name", "stars"]

Checking a global permission

actor = Value("User", "alice")
result = (
oso
.build_query(("has_permission", actor, "create_repository"))
.evaluate()
)
# => true if alice has the global "create_repository" permission

Fetching authorized actions for a collection of resources

repos = ["anvil", "acme"]
actor = Value("User", "alice")
action = typed_var("String")
repo = typed_var("Repository")
results = (
oso
.build_query(("allow", actor, action, repo))
.in_(repo, repos)
.evaluate({repo: action})
)
# => Returns a dict mapping the given repos to the actions alice can perform on those repos- eg.
# { "anvil": ["read"], "acme": ["read", "write"] }

Filtering out unauthorized resources from a collection

repos = ["anvil", "acme"]
actor = Value("User", "bob")
repo = typed_var("Repository")
results = (
oso
.build_query(("allow", actor, "read", repo))
.in_(repo, repos)
.evaluate(repo)
)
# => Returns the subset of `repos` that bob can read- eg.
# ["anvil"]

Filtering an authorize() query based on a relation

actor = Value("User", "bob")
repo = typed_var("Repository");
org = Value("Org", "coolguys")
results = (
oso
.build_query(("allow", actor, "read", repo))
.and_(("has_relation", repo, "parent", org))
.evaluate(repo)
)
# => Returns the IDs of the repos in the coolguys org that bob can read- eg.
# ["acme", "anvil"]

Learn more about how to query Oso Cloud.

Distributed Check API

The distributed check API allows you to perform authorization using data that's distributed across Oso Cloud and your own database. When you instantiate the Oso Cloud client, provide the path to the YAML file that specifies how to resolve facts in your database.


oso = Oso(
...
data_bindings="path/to/data_bindings.yaml",
)

For more information, see the guide on filtering lists with decentralized data.

List authorized resources with distributed data: oso.list_local(actor, action, resource_type, column)

Fetches a filter that can be applied to a database query to return just the resources on which an actor can perform an action. Example with SQLAlchemy:


from sqlalchemy import select, text
alice = Value("User", "alice")
authorized_issues = session.scalars(
select(Issues)
.filter(text(oso.list_local(alice, "read", "Issue", "id")))
).all()

You may use the SQLAlchemy query builder (opens in a new tab) to combine this authorization filter with other things such as ordering and pagination.

Check a permission with distributed data: oso.authorize_local(actor, action, resource)

Fetches a query that can be run against your database to determine whether an actor can perform an action on a resource. Example with SQLAlchemy:


from sqlalchemy import text
alice = Value("User", "alice")
swage_issue = Value("Issue", "swage")
query = oso.authorize_local(alice, "read", swage_issue)
authorized = session.execute(text(query)).scalar()
if not authorized:
raise Exception("Action is not allowed")

The query will return a single boolean value. The user is authorized to perform the action if that boolean value is true.

List authorized actions with distributed data: oso.actions_local(actor, resource)

Fetches a query that can be run against your database to fetch the actions an actor can perform on a resource. Example with SQLAlchemy:


from sqlalchemy import select, text
alice = Value("User", "alice")
swage_issue = Value("Issue", "swage")
query = oso.actions_local(alice, swage_issue)
actions = session.execute(text(query)).scalars()

Policy API

Update the active policy: oso.policy(policy)

Updates the policy in Oso Cloud. The string passed into this method should be written in Polar. Example:


oso.policy("actor User {}")

This command will run any tests defined in your policy. If one or more of these tests fail, your policy will not be updated.

Get policy metadata: oso.get_policy_metadata()

Returns metadata about the currently active policy. Example:


metadata = oso.get_policy_metadata()
print(metadata.resources.keys())
# returns ["Organization", "User", "global"]
print(metadata.resources["Organization"].roles)
# returns ["admin", "member"]

See the Policy Metadata guide for more information on use cases.

Talk to an Oso Engineer

If you'd like to learn more about using Oso Cloud in your app or have any questions about this guide, schedule a 1x1 with an Oso engineer. We're happy to help.

Get started with Oso Cloud →

Migrating to v2