AI Code Assistant Learning from GitHub Repository - n8n Workflow

• Developers and DevOps Teams: Who need fast, context-aware answers about large codebases.


• Technical Writers: Seeking an automated way to query documentation files stored in GitHub.


• Automation Engineers: Looking for advanced examples of using the n8n AI Agent n8n node and RAG patterns.


• Users exploring n8n templates: Specifically those focused on complex AI integrations.

Managing vast amounts of source code or documentation often makes finding specific technical details challenging. This innovative n8n workflow solves this problem by turning your GitHub repository into a searchable knowledge base for an AI Agent.

This robust n8n template separates the process into two key phases: a synchronization flow (indexing the data) and a real-time conversational flow (answering questions). It demonstrates how seamlessly n8n handles large binary data, text splitting, embedding generation, and vector storage, all within a single, coherent n8n workflow. By using this n8n node setup, developers can query their codebase conversationally, drastically reducing search time and enhancing productivity.

This comprehensive n8n workflow operates in two distinct stages:

Stage 1: Data Synchronization (Indexing)

1. Indexing Initiation: The process starts with the Sync Data n8n trigger (a Manual Trigger node).


2. Configuration: The Config n8n node defines critical parameters like repoowner, reponame, and subpath within the GitHub repository.


3. File Retrieval: The GitHub List files n8n node uses these parameters to identify all relevant files in the designated path.


4. Download and Load: The HTTP Request Get File n8n node downloads the actual content of each file.


5. Preprocessing: The Default Data Loader handles the binary data, and the Recursive Character Text Splitter chops the content into smaller, manageable chunks (with 100 characters of overlap) suitable for vector embedding.


6. Embedding and Storage: The Embeddings OpenAI n8n node calculates the vector embeddings for these chunks. Finally, the Simple Vector Store1 n8n node inserts the vector data into an in-memory knowledge base keyed as source-code.

Stage 2: Conversational AI Agent

1. Conversation Start: The workflow is triggered by the When chat message received n8n trigger when a user asks a question.


2. AI Agent Activation: The core AI Agent n8n node takes the query and leverages the OpenAI Chat Model1 for reasoning.


3. Memory and Context: The Window Buffer Memory n8n node maintains conversation history.


4. RAG Lookup: The agent uses the specialized Vector Store Tool n8n node, named projectsource_tool, to perform a similarity search (top K=5) against the indexed knowledge base (Simple Vector Store). This retrieval step is crucial for contextual grounding.


5. Response Generation: The agent receives the relevant context snippets from the repository and generates a precise, technical answer to the user's question, completing the complex n8n workflow.

To deploy this powerful n8n workflow using these n8n templates, follow these steps:

1. Import the JSON: Copy the provided JSON data and import it into your n8n instance via the Workflows section (Import from JSON).


2. Credential Setup: Ensure you have the necessary credentials configured:

1. Configure Parameters: Open the Config n8n node and update the parameters according to your repository:

1. Initial Sync: Run the Sync Data n8n trigger once to populate the Vector Store with your GitHub content before using the chat agent.

Sync Data (Manual Trigger): The starting n8n trigger for the indexing side of the n8n workflow, used to manually refresh the knowledge base.
Config (Set): A foundational n8n node used to dynamically set environment variables (repoowner, reponame, subpath) used throughout the indexing flow.
List files (GitHub): This n8n node retrieves a list of files from the configured GitHub repository path, providing file details needed for downloading.
Get File (HTTP Request): Downloads the actual content of the files using the downloadurl provided by the GitHub n8n node, preparing the data for embedding.
Recursive Character Text Splitter: A key RAG preprocessing n8n node that efficiently splits large files into smaller chunks, optimizing them for embedding and vector search. Configured with a chunk overlap of 100.
Embeddings OpenAI / Embeddings OpenAI1: Two instances of this n8n node, responsible for converting text chunks and incoming queries into high-dimensional numerical vectors using the OpenAI service.
Simple Vector Store / Simple Vector Store1 (Vector Store In Memory): These n8n nodes maintain the RAG knowledge base using the key source-code. One is set to insert mode (for indexing) and the other is set to default (for search retrieval).
When chat message received (Chat Trigger): The initiating n8n trigger for the AI Agent side, listening for user interactions in the n8n chat interface.
AI Agent (LangChain Agent): The primary brain of the n8n workflow. Configured with a system message guiding it to use the projectsourcetool for technical questions based on the source code.
Vector Store Tool: An n8n node that defines how the AI Agent interacts with the knowledge base, specifically configured to retrieve the top 5 most relevant document chunks (topK=5).