cobysy

LLM Wiki

A pattern for building personal knowledge bases using LLMs.

This is an idea file, it is designed to be copy pasted to your own LLM Agent (e.g. OpenAI Codex, Claude Code, OpenCode / Pi, or etc.). Its goal is to communicate the high level idea, but your agent will build out the specifics in collaboration with you.

The core idea

Most people's experience with LLMs and documents looks like RAG: you upload a collection of files, the LLM retrieves relevant chunks at query time, and generates an answer. This works, but the LLM is rediscovering knowledge from scratch on every question. There's no accumulation. Ask a subtle question that requires synthesizing five documents, and the LLM has to find and piece together the relevant fragments every time. Nothing is built up. NotebookLM, ChatGPT file uploads, and most RAG systems work this way.

cobysy

public interface IFreezable
{
    bool IsFrozen { get; }
    void Freeze();
    void Unfreeze();
    void DeepFreeze();
    void DeepUnfreeze();
}

public abstract class FreezableBase : IFreezable

cobysy

// HashUtils to compute Hash on a complex document

var doc = new Document
{
    Title = "Hash Test",
    Links =
    [
        new() { Url = "https://b.com", Text = "B" },
        new() { Url = "https://a.com", Text = "A" }
    ]

cobysy

            // see https://learn.microsoft.com/en-us/rest/api/aiservices/document-models/analyze-document-from-stream?view=rest-aiservices-v4.0%20(2024-11-30)&tabs=HTTP   
            // for parameters

            var analyzeDocumentOptions = new AnalyzeDocumentOptions(
                "prebuilt-layout",
                binaryData)
            {
                OutputContentFormat = DocumentContentFormat.Markdown
            };

cobysy

public sealed class AzureCliCredentialProxy(IHttpClientFactory httpClientFactory) 
    : DefaultAzureCredential
{
    public override AccessToken GetToken(
        TokenRequestContext requestContext,
        CancellationToken cancellationToken = default)
    {
        return this.GetTokenAsync(
            requestContext,
            cancellationToken).GetAwaiter().GetResult();

cobysy

root = true

[*]
charset = utf-8
end_of_line = lf
indent_style = space
insert_final_newline = true

[*.yml]
indent_size = 2

cobysy

Setup

To begin with, we're going to fetch a copy of the source repo. We're potentially going to be making changes in it that we don't want accidentally pushed back to the origin, so it's much safer to use a copy

Clone the source repo into a directory called source_repo and then cd into it:

git clone https://example.invalid/myrepo.git source_repo
cd source_repo

cobysy

public static class QueryTableHintInterceptorExtensions
{
    public static IQueryable<T> WithTableHint<T>(
        this DbSet<T> source,
        string? hints) where T : class
    {
        if (hints == null)
        {
            return source;
        }

cobysy

 var result = await PublicClientApplicationBuilder.CreateWithApplicationOptions(options: new()
            {
                ClientId = "8bxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx",
                TenantId = "1exxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx"
            })
            .WithRedirectUri(redirectUri: "http://localhost")
            .Build()
            .AcquireTokenInteractive(scopes: "api://7fxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx/some.web.api openid profile offline_access".Split(separator: ' '))
            .WithAccount(account: PublicClientApplication.OperatingSystemAccount)
            .WithPrompt(prompt: Prompt.NoPrompt)

cobysy

git init 

git add .

git commit -m "Initial commit"

git remote add origin git@github.com:cobysy/somerepo.git

git push --force origin master:gh-pages