Multimodal

[Iamsdt]

This pull request introduces a comprehensive media processing and configuration system to the API layer, enabling document extraction, flexible media handling, and multimodal message preprocessing. It adds new modules for document extraction and pipeline orchestration, environment-based media configuration, and integrates media services into the application container. Additionally, it improves the handling of multimodal messages by resolving file references before graph execution, and refines error handling for the checkpointer service.

Media Processing and Extraction:

• Added a new media module with DocumentExtractor (using textxtract) and DocumentPipeline for orchestrating document extraction and handling policies (extract_text, pass_raw, skip). (agentflow_cli/media/__init__.py, agentflow_cli/media/extractor.py, agentflow_cli/media/pipeline.py) [1] [2] [3]
• Introduced compatibility helpers for upstream enum changes to ensure stable document handling modes. (agentflow_cli/media/_compat.py)

Media Configuration and Service Integration:

• Added environment-driven media configuration via MediaSettings and get_media_settings, supporting local, memory, cloud, and Postgres storage. (agentflow_cli/src/app/core/config/media_settings.py)
• Wired the media service and settings into the application dependency container for use throughout the API. (agentflow_cli/src/app/loader.py)

Multimodal Message Preprocessing:

• Implemented a preprocess_multimodal_messages utility that resolves file_id references in multimodal messages, replacing them with cached text or agentflow media URLs before graph execution. (agentflow_cli/src/app/routers/graph/services/multimodal_preprocessor.py, agentflow_cli/src/app/routers/graph/services/graph_service.py) [1] [2] [3] [4]

Error Handling Improvements:

• Added a custom CheckpointerUnavailableError for clearer error reporting when the checkpointer service is not configured. (agentflow_cli/src/app/routers/checkpointer/services/checkpointer_service.py) [1] [2]

Networking Utility Enhancement:

• Improved host normalization in the API command utilities to handle unspecified IP addresses more robustly. (agentflow_cli/cli/commands/api.py) [1] [2]