Today, we are excited to announce a powerful new capability on the Giselle platform: the GitHub Vector Store Node and Query Node. This new feature set marks a significant milestone, enabling you to build sophisticated Retrieval-Augmented Generation (RAG) systems using the content of any public or private GitHub repository—all within our no-code, visual interface.
Our mission has always been to empower you to create powerful AI agents with ease. With these new nodes, you can now transform entire codebases into searchable knowledge sources, allowing your AI agents to reason about, explain, and generate code with a deep understanding of the repository's context.
Node-by-node highlights
Item | GitHub Vector Store Node | Query Node |
---|---|---|
Role | Indexes a GitHub repository (code, docs) into a searchable vector store. | Retrieves relevant information from a specified vector store based on a query (can be direct text or output from other nodes). |
Key Process | Fetches repository content, splits it into chunks, generates embeddings, and stores them for retrieval. | Takes a text query (written directly or received from other nodes), finds the most similar chunks from the vector store, and outputs them as context. |
Ideal scenarios | Creating a persistent, searchable knowledge base from a single source of truth—your codebase. | Powering the "Retrieval" part of a RAG workflow; finding context for AI models to answer questions or perform tasks. |
Giselle Plan | Pro & Free | Pro & Free |
When to use the GitHub Vector Store?
Choose this powerful combination when you need to build AI agents that can:
- Answer Codebase Questions: Create an AI assistant that can answer specific questions about functions, classes, or architectural patterns within your repository (e.g., "How is user authentication handled in this project?").
- Automate Documentation: Generate technical documentation, READMEs, or onboarding guides by letting an AI model analyze the entire codebase.
- Perform Cross-File Analysis: Understand and refactor code that spans multiple files and directories by providing the agent with relevant context from across the repository.
- Accelerate Developer Onboarding: Build a helper bot that new team members can query to quickly get up to speed on a complex project.
The Power of Query Node's Simple Design
The Query Node is designed as a simple yet powerful building block—it receives natural language queries, searches one or more vector stores, and returns results. This intentionally minimalist design unlocks remarkable flexibility. While you can certainly write queries directly into the Query Node for straightforward use cases, its true power emerges when it receives dynamically generated queries from other nodes.
This simple design philosophy enables you to implement sophisticated Advanced RAG techniques without writing a single line of code:
1. Query Rewriting (DMQR-RAG)
Transform vague user questions into multiple precise queries. A Text Generation Node can expand "latest AI trends" into targeted searches like "2024 artificial intelligence breakthroughs," "recent machine learning advances," and "emerging AI applications."
2. HyDE (Hypothetical Document Embeddings) (Gao et al., 2022)
Generate hypothetical "ideal answers" first, then find similar real documents. Connect a Text Generation Node to create synthetic documents, then feed them to the Query Node for similarity search.
3. Sequential Multi-Stage Retrieval (MultiHop-RAG)
Build complex answers step by step. Chain multiple Query Nodes together, with Text Generation Nodes synthesizing context between each retrieval stage.
4. Parallel Retrieval with Synthesis (Speculative RAG)
Search multiple perspectives simultaneously. Run multiple Query Nodes in parallel, then merge results with a Text Generation Node for comprehensive answers.
By treating the Query Node as a composable building block rather than a monolithic solution, Giselle empowers you to construct sophisticated RAG pipelines that rival custom-coded implementations—all through visual, no-code workflows.
Why pair them with Giselle?
1. Seamless No-Code RAG Pipeline
Visually construct a complete RAG workflow by simply connecting nodes. Chain the GitHub Vector Store Node to create your knowledge base, use the Query Node to retrieve relevant context, and pipe that context directly into a Generator Node (like Claude 4 Opus or Gemini 2.5 Pro) to get intelligent, context-aware responses.
2. Leverage State-of-the-Art Coding Models
Combine the retrieved code snippets with the world's most advanced coding models available on Giselle. By providing precise, in-context examples from your own repository, you enable models like Anthropic's Claude 4 Opus to deliver highly accurate code analysis, generation, and refactoring suggestions.
3. Automated and Persistent Knowledge
Once a vector store is created, it can be reused across multiple workflows and agents. Giselle handles the complexity of fetching, chunking, and embedding in the background. Simply connect your GitHub account, install the Giselle GitHub App to your organization or account, and select your repository.
4. Accessibility for Complex Tasks
Building a code-aware RAG system typically requires significant engineering effort. Giselle abstracts this complexity away, making it accessible to all builders. You can now focus on designing your agent's logic, not on managing vector databases or embedding pipelines.
Get started now
- Log in — use GitHub, Google, or email at studio.giselles.ai.
- Connect GitHub & Install App — Connect your GitHub account and install the Giselle GitHub App on your account or organization.
- Create or Open an App — Navigate to your Apps page and create a new app or open an existing one.
- Add GitHub Vector Store Node — From the toolbar, click the "Source" icon and select "GitHub Vector Store" to add it to your canvas.
- Configure the Repository — Select the node (it will show "REQUIRES SETUP"). Click "Select a repository" and choose from your repositories with the Giselle GitHub App installed.
- Add Query Node — Place a Query Node on the canvas and connect the GitHub Vector Store Node's output to the Query Node's input.
- Connect to a Generator — Pipe the Query Node's output into a Generator Node (e.g.,
claude-4-opus-20250514
). - Write Query & Run — Add your query to the Query Node, configure the Generator's prompt, and run the workflow to get code-aware answers from your repository.