Voyageai Cli

Full RAG query: embeds the question with Voyage AI, runs vector search against MongoDB Atlas, and reranks results. Use this when you need to answer a question using the knowledge base.

Raw vector similarity search without reranking. Faster than vai_query but results are ordered by vector distance only. Use for exploratory searches or when you plan to rerank separately.

Rerank documents against a query using Voyage AI reranker. Takes a query and candidate documents, returns them reordered by relevance. Use when you have documents from another source and want to order them by relevance.

Embed text using a Voyage AI model and return the vector representation. Use when you need the raw embedding vector for custom similarity logic, storing in another system, or debugging.

Compare two texts semantically by embedding both and computing cosine similarity. Returns a score from -1 (opposite) to 1 (identical). Use for duplicate detection, relevance checking, or topic comparison.

Generate multimodal embeddings for text, images, and/or video using voyage-multimodal-3.5. Accepts base64 data URLs for media. At least one of text, image, or video must be provided. Supports combining multiple content types in a single embedding.

List available MongoDB collections with document counts and vector index information. Use at the start of a task to discover which knowledge bases exist, or when the user mentions a topic and you need to find the right collection.

List available Voyage AI models with capabilities, benchmarks, and pricing. Use when selecting a model for embedding or reranking, or when the user asks about model tradeoffs.

List all available educational topics with summaries. Call this FIRST to discover what topics vai can explain — covers embeddings, vector search, RAG, reranking, model selection, multimodal, code generation, and more. Then use vai_explain to dive deep into any topic.

Get a detailed explanation of a topic. Covers embeddings, vector search, RAG, MoE architecture, shared space, quantization, multimodal, reranking, and more. If the exact topic isn't found, suggests similar topics. Use vai_topics first to browse available topics.

Estimate costs for Voyage AI embedding and query operations at various scales. Use when planning ingestion, budgeting, or comparing model costs.

Add a document to a collection: chunks the text, embeds each chunk with Voyage AI, and stores them in MongoDB Atlas. Use when the user provides new content to add to the knowledge base.

Index a workspace/codebase for semantic code search. Recursively finds files, chunks content, generates embeddings, and stores in MongoDB. Use this to build a searchable knowledge base from a codebase.

Semantic code search across an indexed codebase. Finds code snippets, functions, and documentation semantically related to your query. Use for understanding unfamiliar codebases or finding relevant code.

Get contextual explanation for code by finding relevant documentation and examples in the indexed knowledge base. Useful for understanding what code does or finding usage examples.

Index a codebase for semantic code search. Accepts a local directory path or a GitHub repository URL. Uses voyage-code-3 by default for code-optimized embeddings. Supports incremental refresh — only re-embeds files that changed since last indexing. Creates a MongoDB Atlas vector search index automatically.

Semantic code search across an indexed codebase. Finds functions, classes, modules, and documentation semantically related to your natural language query. Uses voyage-code-3 by default. Supports filtering by programming language and content category. Results include file paths, line numbers, symbols, and relevance scores.

Full RAG query against an indexed codebase. Embeds your question, performs vector search, reranks results, and returns the most relevant code with context. Optimized for answering questions like "how does X work" or "where is Y implemented". Always reranks for best quality.

Find code semantically similar to a given snippet. Paste in a function, class, or code block and find related implementations across indexed codebases. Useful for finding duplicates, alternative implementations, or understanding patterns. Uses voyage-code-3 which understands both code structure and intent.

Check the status of a code search index. Shows file count, chunk count, languages indexed, last indexing time, and vector search index health. Use this before searching to verify the index is ready, or to decide if a refresh is needed.

Generate a complete, executable vai workflow JSON from a natural language description. Returns the workflow definition, validation results, and execution plan. The generated workflow uses template expressions for step inputs and follows all vai workflow conventions.

Validate a vai workflow JSON definition. Checks for structural errors, unknown tools, circular dependencies, and missing references. Returns validation errors, warnings, execution plan layers, and the dependency graph.