Query and manage Weaviate vector database collections

Provides scripted CLI access to a Weaviate vector database across three categories of operations. Search and query: Query Agent "Ask" mode for a direct, synthesized answer with source citations (collection name plus object ID); Query Agent "Search" mode for browsing raw objects across collections; Hybrid Search as the default general-purpose choice balancing semantic and exact-keyword matching; Semantic Search for conceptually similar content regardless of wording; and Keyword Search for exact terms, IDs, or SKUs. Collection management: list_collections.py to discover what exists, recommended as the first step before any other operation; get_collection.py to inspect a collection's schema, properties, vectorizer, replication factor, and multi-tenancy status; explore_collection.py to see data distribution, top values, and actual content before querying; and create_collection.py for new schemas, deliberately omitting a vectorizer unless the user explicitly requests one, since the default text2vec_weaviate applies otherwise. Data operations: fetch_filter.py for precise ID-based or strictly-filtered retrieval, import.py for loading CSV, JSON, JSONL, or PDF files into a collection (PDF imports auto-create the collection), and example_data.py for generating toy data when nothing else is available.

uv run scripts/list_collections.py

Use it to inspect Weaviate collections, schemas, or data distribution, run any of the five search modes, or import data into a Weaviate-backed workflow. It requires a reachable instance and valid credentials (WEAVIATE_URL, WEAVIATE_API_KEY) before anything live can succeed, and for a user with no instance yet, points to Weaviate Cloud's free sandbox signup rather than assuming self-hosting. External inference-provider keys - OpenAI, Cohere, Anthropic, AWS, Mistral, and eleven others, each with its own env var and forwarded header - are auto-detected but never forwarded automatically; they must be explicitly allowlisted via WEAVIATE_PROVIDER_KEYS as a comma-separated list, and only the providers a given collection's configuration actually uses should be set. Its own limitations note that data import, collection creation, and Query Agent operations can change or expose user data, so the target instance and collection must be confirmed before running any script, and that the included scripts are Weaviate-focused only - they don't substitute for broader data-governance, backup, or production-migration procedures.

Input is a collection name plus operation-specific parameters - a search query, filter criteria, a file path for import, or a property-definition list for collection creation. Output is either a Markdown table, the default and recommended format, or JSON via a --json flag, consistent across every script. The recommended workflow sequence: list collections to see what exists, optionally generate example data if nothing is available, inspect a collection's schema and explore its actual data, create a collection with an explicit property list before importing a CSV/JSON/JSONL file (skipped for PDF, which auto-creates its own collection), then import, then choose the right search script for the task - Ask for cited answers across multiple collections, Search for raw objects, Hybrid as the default, Semantic for conceptual matches, Keyword for exact terms. Named error conditions map directly to fixes: a missing WEAVIATE_URL means the env var isn't set, "Collection not found" means running list_collections.py to see what actually exists, and an authentication error means checking API keys for both Weaviate itself and any configured vectorizer provider.

Built on the official Weaviate Python client and its vector-database instance, self-hosted or Weaviate Cloud, with pluggable external inference providers - Anthropic, OpenAI, Azure OpenAI, Cohere, Mistral, NVIDIA, Voyage AI, xAI, HuggingFace, Jina AI, Google Vertex AI/AI Studio, AWS, Anyscale, Databricks, and Friendli - each authenticated via its own API key forwarded as a distinct HTTP header to Weaviate.

Developers building a Weaviate-backed search or RAG workflow who need to inspect collections, choose the right search mode for a query, or import data via scripted, output-format-consistent commands rather than hand-writing Weaviate client calls for each operation.