Retriever (Agentflow Node)

The Retriever node in InnoSynth-Forjinn's Agentflow is a specialized component designed to fetch relevant information from your configured document stores (vector databases) based on a given query. This node is a cornerstone for implementing Retrieval Augmented Generation (RAG) patterns, allowing your AI to access and utilize external knowledge bases.

Purpose

The Retriever node is ideal for:

• Knowledge Retrieval: Fetching specific documents or chunks of information from a large corpus.
• Contextual Grounding: Providing relevant context to LLMs, enabling them to generate more accurate and informed responses.
• RAG Applications: Building AI applications that can answer questions based on your proprietary data.
• Dynamic Information Access: Allowing agents to dynamically search and retrieve information as needed during a conversation.

Appearance on Canvas

(Image: Placeholder for the Retriever node icon/appearance on the canvas)

Configuration Parameters

The Retriever node requires you to specify the document stores to query and the query itself.

1. Knowledge (Document Stores) (Required)

• Label: Knowledge (Document Stores)
• Type: array of objects
• Description: Select one or more document stores (vector databases) from which to retrieve information. These document stores must have been upserted (populated with data) in advance.
  • Document Store: Choose an available document store.

2. Retriever Query (Required)

• Label: Retriever Query
• Type: string
• Placeholder: Enter your query here
• Description: The query or question that the retriever will use to search the selected document stores. This can be a static string or a dynamic value from a preceding node (e.g., {{userQuestion}}).

3. Output Format (Required)

• Label: Output Format
• Type: options
• Default: text
• Options: Text, Text with Metadata
• Description: Determines the format of the retrieved information in the node's output.
  • Text: Returns only the textual content of the retrieved documents, concatenated.
  • Text with Metadata: Returns a JSON string containing the full document objects, including their content and any associated metadata.

4. Update Flow State (Optional)

• Label: Update Flow State
• Type: array of objects
• Description: Allows you to update the runtime state of the workflow during the retriever's execution.
  • Key: The key of the state variable to update.
  • Value: The new value for the state variable. Supports variables and the output of the retriever ({{ output }}).

Inputs & Outputs

• Input: Typically receives a query string or a trigger to initiate the retrieval process.
• Output: Produces the retrieved documents in the specified Output Format. This output can then be passed to an LLM for summarization, question answering, or further processing.

Example Usage: Answering Questions from a Knowledge Base

Let's build a chatflow that answers user questions using a private knowledge base.

1. Connect a Start node to the input of a Retriever node.
2. Configure Knowledge (Document Stores): Select your pre-populated document store (e.g., "Company Policies Document Store").
3. Configure Retriever Query: Set this to {{question}} (assuming the user's query is passed as question).
4. Configure Output Format: Text (to get just the relevant text).
5. Connect Retriever output to an LLM node.
6. Configure LLM:
   • Set a system message: "You are an assistant that answers questions based on the provided context. If the answer is not in the context, state that you don't know."
   • Set the input to include the retrieved documents: Context: {{output}}\nQuestion: {{question}}
7. Connect LLM output to a Chat Output node.

Now, when a user asks a question, the Retriever node will fetch relevant information from your document store, and the LLM will use that information to formulate an answer.