vector-db（矢量数据库）

vector-db

Purpose

This skill enables the management of vector databases for storing, indexing, and querying high-dimensional vectors, optimizing AI/ML workflows for tasks like similarity searches and embeddings.

When to Use

Use this skill for AI/ML applications requiring fast vector similarity queries, such as building recommendation engines, semantic search in NLP, or image retrieval systems. Apply it when dealing with large-scale vector data (e.g., embeddings from models like BERT) to avoid brute-force comparisons.

Key Capabilities

• Store vectors with metadata and perform efficient nearest-neighbor searches using indexes.
• Support distance metrics like cosine, Euclidean, and dot product for similarity calculations.
• Handle vector dimensions up to 2048 and scale to millions of entries.
• Integrate with embedding models for real-time vector generation and querying.

Usage Patterns

Invoke this skill via CLI for quick operations or through API calls in code. Always set the environment variable $VECTOR_DB_API_KEY for authentication before use. For CLI, prefix commands with vector-db and use JSON config files for complex setups (e.g., config.json with {"dimension": 768, "metric": "cosine"}). In code, use HTTP requests to the API endpoint, ensuring error checking on responses. Pattern: First, create an index; then, insert vectors; finally, query them.

Common Commands/API

Use the CLI tool vector-db or the API at https://api.openclaw.com/vector-db/v1. Authentication requires $VECTOR_DB_API_KEY in headers.

• CLI Command: Create an index vector-db create index --name myindex --dimension 768 --metric cosine --file config.json This initializes a new index; ensure config.json specifies additional options like shards.
• CLI Command: Insert vectors vector-db insert --index myindex --vectors "[0.1, 0.2, 0.3]" --id vec1 Vectors must be in JSON array format; use --batch flag for multiple inserts.
• API Endpoint: Query vectors POST https://api.openclaw.com/vector-db/v1/indexes/myindex/query Body: {"vector": [0.1, 0.2, 0.3], "top_k": 5} Response: JSON array of nearest neighbors.
• API Endpoint: Delete index DELETE https://api.openclaw.com/vector-db/v1/indexes/myindex Include header: Authorization: Bearer $VECTOR_DB_API_KEY

Config format: Use JSON files like {"index_name": "myindex", "vector_size": 768, "distance": "cosine"} for CLI operations.

Integration Notes

Integrate with AI/ML tools by exporting vectors from models and using this skill for storage. Set $VECTOR_DB_API_KEY in your environment or.env file. For Python integration, use requests library:

import requests
headers = {'Authorization': f'Bearer {os.environ.get("VECTOR_DB_API_KEY")}' }
response = requests.post('https://api.openclaw.com/vector-db/v1/indexes/myindex/insert', json={'vectors': [[0.1, 0.2]]}, headers=headers)

Ensure the API base URL matches your deployment; handle rate limits by adding retries. For clustering with aimlops, link via shared IDs (e.g., use skill ID "vector-db" in workflows).

Error Handling

Common errors include authentication failures (HTTP 401) from missing $VECTOR_DB_API_KEY, invalid vector dimensions (e.g., mismatch with index), or network issues. To handle:

• Check for 401 errors and prompt user to set $VECTOR_DB_API_KEY.
• For invalid inputs, use try-except in code: try: response = requests.post(url, json=data) response.raise_for_status() except requests.exceptions.HTTPError as e: print(f"Error: {e} - Check vector dimensions.")
• CLI errors show as "Error: Invalid metric specified"; fix by verifying command flags. Always validate inputs before sending requests.

Concrete Usage Examples

1. Example: Building a simple search engine First, create an index: vector-db create index --name searchindex --dimension 512. Insert embeddings: vector-db insert --index searchindex --vectors '[[0.5, 0.6], [0.7, 0.8]]' --ids 'doc1,doc2'. Query for similarities: Use API POST to /indexes/searchindex/query with body {"vector": [0.5, 0.6], "top_k": 3}. This pattern is ideal for NLP, e.g., searching similar documents based on embeddings.
2. Example: Image similarity in ML pipeline Generate image embeddings with a model, then store: vector-db insert --index imageindex --vectors '[[0.1, 0.2, 0.3]]' --metadata '{"url": "image1.jpg"}'. Query for similar images: CLI vector-db query --index imageindex --vector [0.1, 0.2, 0.3] --top_k 5. Integrate in code by fetching results and filtering by metadata, useful for recommendation systems.

Graph Relationships

• Connected to cluster: aimlops (e.g., shares data pipelines with data-processing skills).
• Relates to: embedding-generation skills (for vector creation) and query-optimization tools (for enhancing searches).
• Links with: ai skills for ML model integration and ml skills for training data storage.