Weaviate Cookbooks App Blueprints
An index-style skill offering architectural patterns, end-to-end blueprints, and best practices for building full-stack AI applications with Weaviate.
Credentials flow through environment variables rather than hardcoding; the template requires a local, gitignored .env; rules state 'do not modify the user's collection unless requested' and 'fill only keys actually used', so least-privilege and confirmation are visible. Deducted: no rollback mechanism shown, in-session collection reconfiguration for re-ranking is only verbally restricted, the referenced project_setup.md contract content is not in evidence, and data-flow disclosure to Weaviate Cloud is not systematic.
Cookbooks are internally consistent: install sets, env vars and code snippets agree, and async_client.md documents sync/async distinctions and pitfalls well. Deducted: no committed tests or CI evidence covering key paths, failure handling largely defers to 'web search extensively', failure feedback quality is thin; under static review cannot exceed 10.
The description and index clearly enumerate eight scenarios with a precise semantic trigger (user wants to build AI apps with Weaviate), and a non-fit range (WEAVIATE_* omissible for non-Weaviate agents) is stated. Deducted: core function depends entirely on Weaviate Cloud and overseas LLM providers, mainland-China reachability is uncertain and undisclosed, and Chinese-language support is not addressed.
Good layered IA (index → shared contracts → cookbooks), centralized environment requirements with a canonical .env template, README install and security notes, BSD-3-Clause license present. Deducted: no version numbers, changelog or explicit maintenance ownership; referenced project_setup.md content absent from evidence; Node v25.3.0+ requirement is unusual and unexplained.
The skill offers reusable architectural patterns and best-practice blueprints that let an agent scaffold complete applications, with clear marginal value. Deducted: static review cannot verify outputs are directly usable; many code blocks are fragments ('...existing code', 'import client here'), completeness depends on agent improvisation, and no representative output verification exists.
Extensive links to official docs, forums and GitHub issues enable cross-checking; facts and suggestions are largely separated. Deducted: no executed reproduction in static review, no committed test suite or CI coverage evidence, and runtime correctness of snippets is unverifiable; scored 4 under the static cap of 5.
- Core function depends on Weaviate Cloud and overseas LLM providers; mainland-China reachability is undisclosed — verify network access before use.
- This is a static source review with no execution; no reproducible test evidence covers key paths.
- Several code blocks are fragments ('...existing code'); output completeness depends on agent fill-in quality.
- No version numbers or changelog; content may drift as the upstream Weaviate API evolves.
- The referenced project_setup.md content is not in this evidence set; review that contract file yourself before use.
What does this skill do, and when should you use it?
weaviate-cookbooks is a skill within Weaviate's official agent-skills repository that acts as an index of implementation guides for Weaviate-powered applications. It covers seven concrete blueprints: a Query Agent chatbot, a data explorer, multimodal PDF RAG, basic RAG, advanced RAG, a basic DSPy agent, and agentic RAG, plus an optional Next.js frontend guide and an async Python client guide. The skill enforces shared project-setup contracts and environment requirements, and directs users without an instance to a free Weaviate Cloud sandbox. It suits developers who already use or plan to use Weaviate and want to scaffold full-stack apps following official best practices.
The skill is a set of Markdown reference documents that a coding agent reads on demand: it first loads the shared project setup contract and environment requirements, then selects the relevant cookbook reference (e.g., Query Agent Chatbot or Agentic RAG) to generate a complete application. Optional references cover a Next.js frontend and production use of the Weaviate Python async client (connection patterns, lifecycle management, multi-cluster setups). It also directs users without an instance to register a free sandbox via the Weaviate Cloud console.
- A backend developer who wants a full-stack chatbot with streaming and chat history using the Weaviate Query Agent
- A data team that needs a browser for Weaviate data with sorting, keyword search, and a tabular view
- Building multimodal document search over PDFs using Weaviate Embeddings (ModernVBERT/colmodernvbert) with Ollama and Qwen3-VL
- An engineer upgrading from basic RAG to advanced RAG with re-ranking, query decomposition, and query rewriting
- Building a tool-calling agent with DSPy structured outputs, or agentic RAG with hierarchical tools and Weaviate Query Agent
- Adding a Next.js frontend to an existing Weaviate backend
What are this skill's strengths and limitations?
- Officially maintained by Weaviate under BSD-3-Clause, so content tracks the product
- Covers a complete progression from basic RAG to agentic RAG, multimodal, and agents
- Each blueprint embeds best practices: connection management, environment requirements, project structure contract
- Includes a production-oriented async client guide and an optional frontend plan
- Tightly bound to the Weaviate ecosystem; not applicable to other vector databases
- No evidence of a test suite or quality validation for the blueprints
- Multimodal RAG depends on external components (Ollama, Qwen3-VL); local resource requirements are not quantified
- It is undocumented whether the skill can be installed standalone from the collection
How do you install this skill?
Install the whole skill collection via any Agent Skills-compatible client: run npx skills add weaviate/agent-skills (Cursor, Claude Code, Gemini CLI, etc.), or in Claude Code use /plugin marketplace add weaviate/agent-skills then /plugin install weaviate@weaviate-plugins, or git clone https://github.com/weaviate/agent-skills.git and point your agent at the directory. This skill lives at skills/weaviate-cookbooks/ in the repo. Set WEAVIATE_URL and WEAVIATE_API_KEY environment variables before use. Whether a single skill can be installed independently of the collection is not documented.
How do you use this skill?
Describe what you want in a compatible agent, e.g. "Build a chatbot using the Query Agent", "Build a multimodal RAG app for my PDF documents", or "Build an agentic RAG app". The agent reads the skill's index and the matching reference document, then generates code following the project setup contract. If you have no instance, register a free sandbox at console.weaviate.cloud first.
How does this skill compare with similar options?
The sibling weaviate skill in the same repo provides utility functions for directly operating a Weaviate database (creating collections, querying, importing data); this skill focuses on blueprints for generating complete application code — they complement rather than replace each other.