Together Agents Composio

Agents Composio

alph-notebooks/together-cookbook / Agents_Composio.ipynb

Export

Run Notebooks

Contents

No cells yet

Add cells to see them here

Using Together AI with Composio to Send Emails

This notebook demonstrates how to integrate Together AI's language models with Composio's tools to create an AI agent that can send emails. We'll show:

Setting up the required packages and authentication
Configuring the email tool with Composio
Creating a basic email-sending agent
Adding preprocessing to customize email behavior
Executing and handling the email responses

Let's see how we can combine these powerful tools to automate email communication!

Prerequisites

Together AI API key - see here here https://api.together.ai/
Composio API key - see here https://app.composio.dev/developers
Gmail account connected to Composio - https://app.composio.dev/integrations - You can also watch this video to see how integrations with Composio work.

[4]

Gmail Configuration

We need to connect our Gmail account to Composio. This will open a browser window for authentication.

[ ]

Basic Email Tool Call

Here we create a simple integration between Together AI's LLM and Composio's Gmail tool:

Initialize Together AI client
Create Composio toolset
Configure email sending capability
Make an LLM call with the email tool to see if everything is working

[1]

/opt/anaconda3/envs/cookbook/lib/python3.12/site-packages/composio/tools/toolset.py:543: UserWarning: Not verifying connected accounts for apps. Actions may fail when the Agent tries to use them.
  warnings.warn(

{'id': 'nnSQh4S-4yUbBN-925f455aa84c36d8',
, 'object': <ObjectType.ChatCompletion: 'chat.completion'>,
, 'created': 1742914753,
, 'model': 'Qwen/Qwen3-235B-A22B-Instruct-2507-tput',
, 'choices': [{'index': 0,
,   'logprobs': None,
,   'seed': 15668879311493286000,
,   'finish_reason': <FinishReason.ToolCalls: 'tool_calls'>,
,   'message': {'role': <MessageRole.ASSISTANT: 'assistant'>,
,    'content': None,
,    'tool_calls': [{'id': 'call_4rx0ntnagtqyrwml4fwvmhz9',
,      'type': 'function',
,      'function': {'name': 'GMAIL_SEND_EMAIL',
,       'arguments': '{"user_id":"me","recipient_email":"john@mlteam.com","subject":"Meeting Request: Review Upcoming Launch","body":"Hi John,\\n\\nI hope this message finds you well. Could you please let me know a good time for us to meet next week to review the upcoming launch? Your availability is much appreciated.\\n\\nBest regards,\\nAlex","is_html":false}'},
,      'index': 0}]}}],
, 'prompt': [],
, 'usage': {'prompt_tokens': 535,
,  'completion_tokens': 104,
,  'total_tokens': 639}}

Modify Gmail tool schema

We can customize the email behavior using preprocessors and schema processors:

Preprocessor: Modifies the input before sending (e.g., changing recipient email)
Schema Processor: Modifies the tool schema (e.g., removing fields from LLM's view)

[2]

Executing the Email Send

Now we'll execute the email send with our processed tool and handle the response

[3]

[4]

[{'data': {'response_data': {'id': '195cdd11b381f385',
,    'threadId': '195cdd11b381f385',
,    'labelIds': ['UNREAD', 'SENT', 'INBOX']}},
,  'error': None,
,  'successfull': True,
,  'successful': True}]