Our next generation SDKs

In the last few months, we have experienced very rapid growth in usage of our platform. As such, our team has been working hard to radically improve the performance and developer experience of our platform.

A lot of these changes have happened in the background, but we are excited to finally share our new SDKs with you that complement our new infra.

The new API features improved usability, enhanced stability, and better scalability. The SDKs built on top of it simplify the developer experience, making it easier than ever to build useful agents.

What’s new?

A lot of the changes are on the infra side, but from the SDK point of view, here is what you can expect:

• Faster and more reliable tool execution
• A simpler but more opinionated SDK
• Much more intuitive and consistent naming conventions
• A vastly improved TypeScript SDK that is meaningfully more type-safe and has full feature parity with the Python SDK

There aren’t too many new flashy features here (yet) mainly because we wanted to get the bones right — but we feel we have a solid foundation to ship incredible new experiences on top very quickly.

State of the new SDK and what is happening with the old SDKs?

Currently, the new SDKs are in a preview release. These new SDKs come almost fully formed, we do not expect many breaking changes to them but are releasing them in a preview state to get feedback and make necessary changes before locking them in.

As we lock the new SDKs in place, we will deprecate support for the old SDKs. They will continue to work for the foreseeable future but are no longer actively maintained. We will continue to push security updates and fix any critical bugs but will not support any new functionality in them.

We urge you to upgrade to the new SDKs as soon as possible.

Nomenclature

We have updated several key terms in the SDK and API to improve clarity and consistency. The following table summarizes these changes:

Switch to nano IDs from UUIDs

We have transitioned from UUIDs to nano IDs throughout the platform for the following reasons:

• Improved readability: UUIDs are lengthy and difficult to read
• Better usability: Easier to copy with a single double-click
• Better organization: Nano IDs allow us to distinguish between different resource types through prefixes

Note: Nano IDs are short, unique, and prefixed to indicate the resource type.

SDK Changes

Upgrade to the latest SDK version using the appropriate package manager:

1
pip install -U composio

Both SDKs now implement proper namespacing for each concept.

User ID scoping

The concept of entity_id has been expanded and renamed to user_id.

All operations are now scoped to a user ID, including:

• Fetching tools
• Initiating connections
• Executing tools
• Managing triggers

This change provides explicit specification of the user for whom the action is being performed. When a user may have multiple accounts (such as work and personal Gmail connections), you can use the more specific connected account ID.

Replacing ToolSets with Providers

We have deprecated “toolsets” in favor of “providers”. This change allows Composio to provide deeper standardization for tool implementation across different frameworks.

Previously, you needed to import and use a framework-specific ComposioToolSet class:

1
from composio_openai import ComposioToolSet, Action, App
2
from openai import OpenAI

The SDK structure is now framework-agnostic and includes the OpenAI provider out of the box:

1
from composio import Composio
2
# from composio_langchain import LangchainProvider
3
4
composio = Composio()
5
# composio = Composio(provider=LangchainProvider())
6
7
tools = composio.tools.get(
8
    user_id="0001",
9
    tools=["LINEAR_CREATE_LINEAR_ISSUE", "GITHUB_CREATE_COMMIT"]
10
)
11
# tools returned is formatted for the provider. by default, OpenAI.

You can now use the same tools across any framework with our unified interface, or create custom toolsets for frameworks we don’t yet support.

Read more about providers in our documentation and explore the complete list of available providers.

Fetching and filtering tools

Previously, you could filter tools by:

• Apps
• Action names (tool names)
• Tags

You could also specify an important flag to retrieve the most important tools:

1
from composio_openai import ComposioToolSet, Action, App
2
from openai import OpenAI
3
4
toolset = ComposioToolSet()
5
client = OpenAI()
6
7
tools = toolset.get_tools(
8
    actions=[Action.GITHUB_GET_THE_AUTHENTICATED_USER], check_connected_accounts=True
9
)
10
11
tools = toolset.get_tools(apps=[App.GITHUB, App.LINEAR, App.SLACK], check_connected_accounts=True)

You can now filter tools by:

• Toolkits
• Tool slugs
• Limit parameter
• Search query

The important flag has been removed. Instead, tools are returned in order of importance by default:

1
from composio import Composio
2
3
composio = Composio()
4
5
user_id = "user@acme.org"
6
7
tools_1 = composio.tools.get(user_id=user_id, toolkits=["GITHUB", "LINEAR"])
8
9
tools_2 = composio.tools.get(user_id=user_id, toolkits=["SLACK"], limit=5)  # Default limit=20
10
11
tools_3 = composio.tools.get(
12
    user_id=user_id,
13
    tools=["GITHUB_CREATE_AN_ISSUE", "GITHUB_CREATE_AN_ISSUE_COMMENT", "GITHUB_CREATE_A_COMMIT"],
14
)
15
16
tools_4 = composio.tools.get(user_id="john", search="hackernews posts")

Fetching raw tool data

To examine the raw schema definition of a tool for understanding input/output parameters or building custom logic around tool definitions, use the following methods:

1
from composio import Composio
2
3
composio = Composio()
4
5
tool = composio.tools.get_raw_composio_tool_by_slug("HACKERNEWS_GET_LATEST_POSTS")
6
7
print(tool.model_dump_json())

Executing tools

Tool execution remains largely unchanged, with user_id now explicitly required.

For agentic frameworks, the tool object returned from tools.get is now the respective framework’s native tool object. Tool call execution is handled by the agentic framework itself.

1
from composio import Composio
2
from openai import OpenAI
3
4
openai_client = OpenAI()
5
composio = Composio()
6
7
tools = composio.tools.get(user_id="user@acme.com", tools=["GITHUB_GET_THE_ZEN_OF_GITHUB"])
8
response = openai_client.chat.completions.create(
9
    model="gpt-4.1",
10
    messages=[{"role": "user", "content": "gimme some zen."}],
11
    tools=tools,
12
)
13
14
result = composio.provider.handle_tool_calls(user_id="user@acme.com", response=response)
15
print(result)

For more information on executing tools for different frameworks, see Replacing ToolSets with Providers.

Tool Modifiers (formerly Tool Processors)

Tool processors have been renamed to tool modifiers and now provide an improved developer experience. The implementation is now available in TypeScript too! (previously Python-only).

1
from composio_openai import ComposioToolSet, Action
2
3
toolset = ComposioToolSet()
4
5
6
def my_schema_processor(schema: dict) -> dict: ...
7
def my_preprocessor(inputs: dict) -> dict: ...
8
def my_postprocessor(result: dict) -> dict: ...
9
10
11
# Get tools with the modified schema
12
processed_tools = toolset.get_tools(
13
    actions=[Action.GMAIL_SEND_EMAIL],
14
    processors={
15
        # Applied BEFORE the LLM sees the schema
16
        "schema": {Action.SOME_ACTION: my_schema_processor},
17
        # Applied BEFORE the tool executes
18
        "pre": {Action.SOME_ACTION: my_preprocessor},
19
        # Applied AFTER the tool executes, BEFORE the result is returned
20
        "post": {Action.SOME_ACTION: my_postprocessor},
21
    },
22
)

The modifiers now leverage language-specific features to provide a more natural developer experience.

While tool processors could previously be applied during SDK initialization, tool fetching, and tool execution, we have restructured them as follows:

• Chat Completion providers: Modifiers are specified and applied during tool execution
• Agentic frameworks: Modifiers are specified and applied during tool fetching

Schema Modifiers

The following example demonstrates schema modifier usage, applicable across all providers:

1

Before Modifiers

The following example shows creating and using a before modifier for a Chat Completion provider. For agentic frameworks, view the complete before modifier documentation:

1

After Modifiers

The following example shows creating and using an after modifier for a Chat Completion provider. For agentic frameworks, view the complete after modifier documentation:

1

Custom Tools

The SDK continues to support custom tools. Creating tools from your methods remains possible. We recommend reviewing the detailed custom tools documentation for more information.

Due to changes in the SDK architecture, creating custom tools that use Composio’s managed authentication has been modified. In the previous SDK, you could create a custom tool as follows:

1
# Python Example using execute_request
2
from composio import action, ComposioToolSet
3
import typing as t
4
5
toolset = ComposioToolSet()
6
7
@action(toolname="github") # Associate with GitHub app for auth
8
def get_github_repo_topics(
9
    owner: t.Annotated[str, "Repository owner username"],
10
    repo: t.Annotated[str, "Repository name"],
11
    execute_request: t.Callable # Injected by Composio
12
) -> dict:
13
    """Gets the topics associated with a specific GitHub repository."""
14
    response_data = execute_request(
15
        endpoint=f"/repos/{owner}/{repo}/topics", # API path relative to base URL
16
        method="GET"
17
    )
18
    if isinstance(response_data, dict):
19
        return {"topics": response_data.get("names", [])}

The execute tool request method handles injection of the appropriate base URL and authentication credentials for the tool:

1
from pydantic import BaseModel, Field
2
from composio import Composio
3
from composio.core.models.custom_tools import ExecuteRequestFn
4
5
6
composio = Composio()
7
8
class GetIssueInfoInput(BaseModel):
9
    issue_number: int = Field(
10
        ...,
11
        description="The number of the issue to get information about",
12
    )
13
14
# function name will be used as slug
15
@composio.tools.custom_tool(toolkit="github")
16
def get_issue_info(
17
    request: GetIssueInfoInput,
18
    execute_request: ExecuteRequestFn,
19
    auth_credentials: dict,
20
) -> dict:
21
    """Get information about a GitHub issue."""
22
    response = execute_request(
23
        endpoint=f"/repos/composiohq/composio/issues/{request.issue_number}",
24
        method="GET",
25
        parameters=[
26
            {
27
                "name": "Accept",
28
                "value": "application/vnd.github.v3+json",
29
                "type": "header",
30
            },
31
            {
32
                "name": "Authorization",
33
                "value": f"Bearer {auth_credentials['access_token']}",
34
                "type": "header",
35
            },
36
        ],
37
    )
38
    return {"data": response.data}

For more information, including executing custom tools and defining custom headers and query parameters, refer to the Custom Tools documentation.

Auth configs (formerly integrations)

Integrations are now called auth configs. While the terminology has changed, the underlying concept remains the same.

Auth configs store the configuration required for authentication with a given toolkit, including OAuth developer credentials, configurable base URLs, and scopes.

Auth configs now use nano IDs instead of UUIDs:

We recommend storing auth config nano IDs in your database for connecting users to the appropriate auth configuration.

For most use cases, you will create auth configs through the dashboard, and this process remains unchanged. Read more about creating auth configs and customizing auth configs.

Creating auth configs programmatically in the previous SDK:

1
from composio_openai import App, ComposioToolSet
2
3
toolset = ComposioToolSet()
4
5
integration = toolset.create_integration(
6
    app=App.GITHUB,
7
    auth_mode="OAUTH2",
8
    use_composio_oauth_app=True,
9
    # For use_composio_oauth_app=False, you can provide your own OAuth app credentials here
10
    # auth_config={
11
    #     "client_id": "123456",
12
    #     "client_secret": "123456"
13
    # }
14
15
)

Creating auth configs programmatically in the current SDK:

1
from composio import Composio
2
3
composio = Composio()
4
5
# Use composio managed auth
6
auth_config = composio.auth_configs.create(
7
    toolkit="notion",
8
    options={
9
        "type": "use_composio_managed_auth",
10
        # "type": "use_custom_auth",
11
        # "auth_scheme": "OAUTH2",
12
        # "credentials": {
13
        #     "client_id": "1234567890",
14
        #     "client_secret": "1234567890",
15
        #     "oauth_redirect_uri": "https://backend.composio.dev/api/v3/toolkits/auth/callback",

For using custom authentication credentials, refer to the Programmatic Auth Configs documentation.

Connected accounts / User IDs

The primary change in connected accounts and user IDs is that user IDs are now a more prominent concept compared to entities in previous versions.

We have simplified the process of connecting a user to a toolkit. Instead of multiple methods and parameters for initiating a connection, both the SDK and API now require only a user_id and auth_config_id to initiate a connection.

This approach is more explicit and works well with the ability for developers to have multiple auth configs for a given toolkit.

Connected accounts now use nano IDs instead of UUIDs:

Previously, you might have initiated a connection like this:

1
from composio_openai import ComposioToolSet
2
3
toolset = ComposioToolSet()
4
user_id = "your_user_unique_id"
5
google_integration_id = "0000-0000"
6
7
entity = toolset.get_entity(id=user_id)
8
9
try:
10
    print(f"Initiating OAuth connection for entity {entity.id}...")
11
    connection_request = toolset.initiate_connection(
12
        integration_id=google_integration_id,
13
        entity_id=user_id,
14
        # Optionally add: redirect_url="https://yourapp.com/final-destination"
15
        # if you want user sent somewhere specific *after* Composio finishes.
16
    )
17
18
    # Check if a redirect URL was provided (expected for OAuth)
19
    if connection_request.redirectUrl:
20
        print(f"Received redirect URL: {connection_request.redirectUrl}")
21
    else:
22
        print("Error: Expected a redirectUrl for OAuth flow but didn't receive one.")
23
24
25
except Exception as e:
26
    print(f"Error initiating connection: {e}")

The current process for initiating a connection is as follows:

1
from composio import Composio
2
3
linear_auth_config_id = "ac_1234"
4
user_id = "user@email.com"
5
composio = Composio()
6
7
# Create a new connected account
8
connection_request = composio.connected_accounts.initiate(
9
    user_id=user_id,
10
    auth_config_id=linear_auth_config_id,
11
)
12
print(connection_request.redirect_url)
13
14
# Wait for the connection to be established
15
connected_account = connection_request.wait_for_connection()

Triggers

Composio continues to support listening to application events using triggers through WebSockets and webhooks.

Creating triggers

The process for creating triggers and specifying their configuration has been redesigned for improved clarity and intuitiveness.

Some triggers require configuration, such as repository names for GitHub triggers or channel names for Slack triggers. The process usually follows the pattern of fetching the trigger type and then creating the trigger with the appropriate configuration.

1
from composio import Composio
2
3
composio = Composio()
4
5
user_id = "user@example.com"
6
trigger_config = composio.triggers.get_type("GITHUB_COMMIT_EVENT")
7
print(trigger_config.config)
8
### Trigger Config
9
# {
10
#     "properties": {
11
#         "owner": {
12
#             "description": "Owner of the repository",
13
#             "title": "Owner",
14
#             "type": "string"
15
#         },
16
#         "repo": {
17
#             "description": "Repository name",
18
#             "title": "Repo",
19
#             "type": "string"
20
#         }
21
#     },
22
#     "required": ["owner", "repo"],
23
#     "title": "WebhookConfigSchema",
24
#     "type": "object"
25
26
trigger = composio.triggers.create(
27
    slug="GITHUB_COMMIT_EVENT",
28
    user_id=user_id,
29
    trigger_config={"repo": "composiohq", "owner": "composio"},
30
)
31
print(trigger)

Enabling/Disabling triggers

You can enable or disable triggers through either the SDK or the dashboard. The dashboard process remains unchanged.

Managing triggers with the SDK:

1
# Disable a trigger instance 
2
disabled_instance = composio.triggers.disable(trigger_id="ti_abcd123")
3
print(disabled_instance) 

If needed, the trigger can be enabled again.

1
# Enable a trigger instance
2
enabled_instance = composio.triggers.enable(trigger_id="ti_abcd123")
3
print(enabled_instance)

Listening to triggers

We recommend listening to triggers through webhooks. The following are example routes for Next.js and FastAPI.

For development, you can also listen to triggers through the SDK.

1
from fastapi import FastAPI, Request
2
from typing import Dict, Any
3
import uvicorn
4
import json
5
6
app = FastAPI(title="Webhook Demo")
7
8
@app.post("/webhook")
9
async def webhook_handler(request: Request):
10
    # Get the raw payload
11
    payload = await request.json()
12
    
13
    # Log the received webhook data
14
    print("Received webhook payload:")
15
    print(json.dumps(payload, indent=2))
16
    
17
    # Return a success response
18
    return {"status": "success", "message": "Webhook received"}
19
20
if __name__ == "__main__":
21
    uvicorn.run(app, host="0.0.0.0", port=8000)

🚧 Coming Soon

Local tools

Previously, the Python SDK included local tools. These were tools defined within the SDK and consisted of local shell and code-related tools such as “clipboard”, “sqltool”, and “shelltool”.

This feature is currently in development for both Python and TypeScript SDKs, with newly created tools built for improved agent accuracy.

This feature is currently in development for both Python and TypeScript SDKs.

API Endpoints

The following table lists important API endpoints that have changed. You can use this reference to quickly find the new v3 API endpoint for migration:

Toolkits (formerly Apps)

Tools (formerly Actions)

Auth Configs (formerly Integrations/Connectors)

Connected Accounts (formerly Connections)

Triggers