Subscribe to our Newsletter for Updates and Tips¶

Understanding Semantic Validation with Structured Outputs

Semantic validation uses LLMs to evaluate content against complex, subjective, and contextual criteria that would be difficult to implement with traditional rule-based validation approaches.

As LLMs become increasingly integrated into production systems, ensuring the quality and safety of their outputs is paramount. Traditional validation methods relying on explicit rules can't keep up with the complexity and nuance of natural language. With the release of Instructor's semantic validation capabilities, we now have a powerful way to validate structured outputs against sophisticated criteria.

Announcing Responses API support

We're excited to announce Instructor's integration with OpenAI's new Responses API. This integration brings a more streamlined approach to working with structured outputs from OpenAI models. Let's see what makes this integration special and how it can improve your LLM applications.

Announcing unified provider interface

We are pleased to introduce a significant enhancement to Instructor: the from_provider() function. While Instructor has always focused on providing robust structured outputs, we've observed that many users work with multiple LLM providers. This often involves repetitive setup for each client.

The from_provider() function aims to simplify this process, making it easier to initialize clients and experiment across different models.

This new feature offers a streamlined, string-based method to initialize an Instructor-enhanced client for a variety of popular LLM providers.

Using Anthropic's Web Search with Instructor for Real-Time Data

Anthropic's new web search tool, when combined with Instructor, provides a powerful way to get real-time, structured data from the web. This allows you to build applications that can answer questions and provide information that is up-to-date, going beyond the knowledge cut-off of large language models.

In this post, we'll explore how to use the web_search tool with Instructor to fetch the latest information and structure it into a Pydantic model. Even a simple structure can be very effective for clarity and further processing.

What is MCP

With OpenAI joining Anthropic in supporting the Model Context Protocol (MCP), we're witnessing a unified standard for language models to interact with external systems. This creates exciting opportunities for multi-LLM architectures where specialized AI applications work in parallel-discovering tools, handing off tasks, and accessing powerful capabilities through standardized interfaces.

Instructor Adopts llms.txt: Making Documentation AI-Friendly

We're excited to announce that Instructor now implements the llms.txt specification! You can now find our llms.txt file at python.useinstructor.com/llms.txt. This adoption marks an important step in making our documentation more accessible to AI language models.

Instructor Adopting Cursor Rules

AI-assisted coding is changing how we use version control. Many developers now use what I call "vibe coding" - coding with AI help. This creates new challenges with Git. Today I'll share how we're using Cursor rules in Instructor to solve these problems.

Migrating to uv

Why we migrated to uv

We recently migrated to uv from poetry because we wanted to benefit from it's many features such as

• Easier dependency management with automatic caching built in
• Significantly faster CI/CD compared to poetry, especially when we use the caching functionality provided by the Astral team
• Cargo-style lockfile that makes it easier to adopt new PEP features as they come out

We took around 1-2 days to handle the migration and we're happy with the results. On average, for CI/CD, we've seen a huge speed up for our jobs.

Here are some timings for jobs that I took from our CI/CD runs.

In general I'd say that we saw a ~3x speedup with approximately 67% reduction in time needed for the jobs once we implemented caching for the individual uv github actions.

Extracting Metadata from Images using Structured Extraction

Multimodal Language Models like gpt-4o excel at processing multimodal, enabling us to extract rich, structured metadata from images.

This is particularly valuable in areas like fashion where we can use these capabilities to understand user style preferences from images and even videos. In this post, we'll see how to use instructor to map images to a given product taxonomy so we can recommend similar products for users.

Consistent Stories with GPT-4o

Language Models struggle to generate consistent graphs that have a large number of nodes. Often times, this is because the graph itself is too large for the model to handle. This causes the model to generate inconsistent graphs that have invalid and disconnected nodes among other issues.

In this article, we'll look at how we can get around this limitation by using a two-phase approach to generate complex DAGs with gpt-4o by looking at a simple example of generating a Choose Your Own Adventure story.