LLM Docs

Token-Optimized Minified Popular Docs

14 followers

Token-Optimized Minified Popular Docs

14 followers

Instant access to single file documentation for popular libraries. Optimized for LLM context windows with minified versions. Search, compare, and copy documentation with ease.

Free

Launch tags:Productivity•Developer Tools•Artificial Intelligence

Launch Team

Unblocked AI Code Review — High-signal comments based on your team's context

High-signal comments based on your team's context

Promoted

Maker

📌

Hello Product Hunters,

I've been programming for over 20 years, and in the past year, I've increasingly integrated LLMs into my workflow, currently using Cursor with agents to speed up my development. One challenge I've consistently encountered is that LLMs often struggle with newer or less well-known libraries. They sometimes fail to recognize the most recent APIs, such as those for Polars, Skeleton UI, or Svelte 5. To work around this, I've frequently had to manually copy relevant documentation into the LLM to guide it in understanding the correct usage.

After doing this many times, I realized it would be incredibly useful to have a centralized repository of documentation designed specifically for LLMs. The idea was to offer comprehensive, token-friendly documentation formats that can be directly pasted into an LLM prompt. Since I couldn’t find an existing solution, I decided to build it myself and share it with everyone.

At llm-docs, you’ll find four documentation formats:

Full: A single file containing the complete concatenated documentation.
Minified: The full documentation condensed using state-of-the-art LLM summarization.
Core (up to 50k tokens): The most important parts of the API, selected via an ELO ranking system.
Slim (up to 10k tokens): A compact version featuring only the critical definitions, ideal for token-sensitive applications.

I hope you find llm-docs useful, and I look forward to your feedback and suggestions!

Report

1yr ago

Maker

@masump Each full documentation consists of many individual pieces, and then using SOTA LLM I run the following prompt:

prompt = """
You are a domain expert in programming with the library "{library_name}". Below are two text snippets that are intended to represent API documentation or function/method definitions. Your task is twofold:
Evaluate each snippet for its relevance as "core" documentation for using and programming with the library. In your evaluation, give strong priority to documentation that directly instructs the user on how to use the API. This includes explicit function or method signatures, parameter descriptions, return types, usage examples, or code snippets that clearly demonstrate how to call and use the function. Assign a "relevance_score" as a float on a scale from 1.0 to 10.0, where:
10.0 indicates that the snippet provides detailed, actionable documentation that explicitly explains how to use the function or API.
1.0 indicates that the snippet is generic, high-level, or lacks concrete details necessary for practical implementation.
For each snippet, determine whether it "contains_relevant_docs". Set this value to true only if the snippet includes specific, actionable documentation text—such as explicit function definitions, parameter lists, return value descriptions, or code examples. If the snippet only lists topics, describes high-level applications, or is merely descriptive without direct usage instructions, set this value to false.
Based on these evaluations, decide which snippet is more "core" for understanding and using the library, and which one a typical user is more likely to actually use. Consider that "more_important" should be the snippet that not only is more core in its documentation but also is more directly useful for a user implementing or interacting with the library.
Please output your response strictly as a JSON object in the following format and nothing else:
{{
"A": {{"relevance_score": <number>, "contains_relevant_docs": <true or false>}},
"B": {{"relevance_score": <number>, "contains_relevant_docs": <true or false>}},
"more_important": "<A or B>"
}}
Do not include any additional text or explanation.
Snippet A:
{snippet_a}

Snippet B:
{snippet_b}
"""

Where Snippet A is one piece of documentation and Snippet B is another piece of documentation. This is run many times and once there is sufficient number of comparisons of each snippet we can calculate ELO of individual snippets and then use that as their "importance" to be included in core and slim versions.

Report

1yr ago

Full: A single file containing the complete concatenated documentation.
Minified: The full documentation condensed using state-of-the-art LLM summarization.
Core (up to 50k tokens): The most important parts of the API, selected via an ELO ranking system.
Slim (up to 10k tokens): A compact version featuring only the critical definitions, ideal for token-sensitive applications.

I hope you find llm-docs useful, and I look forward to your feedback and suggestions!

prompt = """ You are a domain expert in programming with the library "{library_name}". Below are two text snippets that are intended to represent API documentation or function/method definitions. Your task is twofold: Evaluate each snippet for its relevance as "core" documentation for using and programming with the library. In your evaluation, give strong priority to documentation that directly instructs the user on how to use the API. This includes explicit function or method signatures, parameter descriptions, return types, usage examples, or code snippets that clearly demonstrate how to call and use the function. Assign a "relevance_score" as a float on a scale from 1.0 to 10.0, where: 10.0 indicates that the snippet provides detailed, actionable documentation that explicitly explains how to use the function or API. 1.0 indicates that the snippet is generic, high-level, or lacks concrete details necessary for practical implementation. For each snippet, determine whether it "contains_relevant_docs". Set this value to true only if the snippet includes specific, actionable documentation text—such as explicit function definitions, parameter lists, return value descriptions, or code examples. If the snippet only lists topics, describes high-level applications, or is merely descriptive without direct usage instructions, set this value to false. Based on these evaluations, decide which snippet is more "core" for understanding and using the library, and which one a typical user is more likely to actually use. Consider that "more_important" should be the snippet that not only is more core in its documentation but also is more directly useful for a user implementing or interacting with the library. Please output your response strictly as a JSON object in the following format and nothing else: {{ "A": {{"relevance_score": <number>, "contains_relevant_docs": <true or false>}}, "B": {{"relevance_score": <number>, "contains_relevant_docs": <true or false>}}, "more_important": "<A or B>" }} Do not include any additional text or explanation. Snippet A: {snippet_a} Snippet B: {snippet_b} """