Conversation Files

Conversation Files gives an agent access to the files shared in the current conversation: documents, images or videos the user has attached. With it, the agent can see which files there are, read their content, summarise them or look for a specific detail inside them.

You connect its Tools output to the agent’s Tools input and choose in the Tools field which actions you allow. The agent decides each time which one to use based on what it is asked.

When to use it

A customer attaches a contract to the chat and asks about a specific clause.
An employee uploads several invoices and asks the assistant for a summary of each.
A user sends a photo of a damaged product and the agent has to describe what it sees.
You want the agent to remember what was said earlier in the same conversation.

Configuration

Field	What it does
Tools (required)	The actions the agent can use. You can pick several. Defaults to `list_conversation_files` only.
Model	The AI model that describes images and videos and writes the summaries. Connect it if you enable reading or summarising files.
Embedding	The embedding model that makes it possible to search by meaning inside the files. Connect it if you enable `search_in_conversation_files`.

Actions available in Tools:

Action	What the agent can do
`list_conversation_files`	See the list of files in the conversation, with their name, type and size.
`get_conversation_file_content`	Read the full content of a file. For an image or a video, it gets a description written by the Model.
`get_conversation_file_metadata`	Look up a file’s details: name, address, type and size.
`get_conversation_file_resume`	Get a summary of a file following instructions the agent writes. Needs Model.
`search_in_conversation_files`	Search for a topic or phrase across all files and get the closest passages. Needs Embedding.
`get_memory_information`	Look up the earlier messages of the conversation.

Output

Tools: connects to an agent’s Tools input and gives it the chosen actions.
Data: the list of files in the conversation, with their identifier, name, address, type and size. It is hidden when you connect the Tools output.

Example

A law firm offers a chat where clients attach documents.

Add Conversation Files and connect its Tools output to the agent’s Tools input.
In Tools, pick list_conversation_files, get_conversation_file_resume and search_in_conversation_files.
Connect an AI model to Model and an embedding model to Embedding.
A client uploads a rental contract and asks “How much notice do I need to give before leaving?”. The agent finds the passage about notice in the document and answers with what it says.

Good practices

Enable only the actions the agent needs: with fewer options it chooses better.
Image and video descriptions depend on the connected Model being able to understand images or video.

Agent: the agent this component connects to.
Conversations: where you see your flows’ conversations.