feat: Add task-native and agent-native multimodality support

[ssam18]

Summary

This PR implements native multimodality support for crewAI Agents and Tasks, addressing feature request #3860.

Changes

New Modules

multimodal/image.py

• Image class for handling images in multiple formats:
  • HTTP(S) URLs via from_url()
  • Local file paths via from_file() with automatic MIME type detection
  • Base64 encoded strings via from_base64()
  • Binary data via from_binary()
  • Runtime placeholders via from_placeholder()
• Automatic source type inference
• Conversion methods: to_data_url(), to_message_content()
• Compatible with OpenAI, Anthropic, and other vision-capable LLM APIs

multimodal/multipart_content.py

• MultipartContent class for managing compound text+image context
• Methods: add_text(), add_image(), to_message_content()
• Utility methods: get_text_only(), has_images()

Integration

agent/core.py and task.py

• Added multipart_context field to both Agent and Task classes
• Field type: list[str | Image] | MultipartContent | None
• Enables multimodal system prompts for agents and multimodal inputs for tasks

Usage Example

from crewai import Agent, Task
from crewai.multimodal import Image, MultipartContent

# Agent with multimodal context
agent = Agent(
    role="Vision Analyst",
    goal="Analyze images and diagrams",
    multipart_context=[
        "You are an expert at analyzing visual content.",
        Image.from_url("https://example.com/reference.png")
    ]
)

# Task with multipart content
content = MultipartContent()
content.add_text("Compare these two architectural diagrams:")
content.add_image(Image.from_file("diagram1.png"))
content.add_image(Image.from_file("diagram2.png"))

task = Task(
    description="Identify differences between the diagrams",
    multipart_context=content,
    agent=agent
)

Testing

Comprehensive test suite included in tests/multimodal/:

• test_image.py: 19 test cases covering all Image class functionality
• test_multipart_content.py: 13 test cases for MultipartContent
• run_tests.py: Custom test runner for environments without pytest
• validate_implementation.py: Syntax and structure validation

All validation checks pass successfully.

Closes

#3860

---

Note

Introduces multimodal support via new Image and MultipartContent classes and integrates them as multipart_context on Agent and Task, with comprehensive tests.

• Multimodal module (crewai/multimodal/):
  • Image: constructors for URL/file/base64/binary/placeholder; conversion to data URLs and LLM message content.
  • MultipartContent: manage mixed text+images; convert to LLM message parts; text-only extraction and helpers.
• Core integration:
  • crewai/agent/core.py: add multipart_context: list[str | Image] | MultipartContent | None to include multimodal context in system prompt.
  • crewai/task.py: add multipart_context: list[str | Image] | MultipartContent | None to enrich task context.
• Tests (tests/multimodal/):
  • Unit tests for Image and MultipartContent; integration checks; standalone runner and validator.

^{Written by Cursor Bugbot for commit 448c815. This will update automatically on new commits. Configure here.}