Inbed

Inbed is a TypeScript library for semantic indexing and retrieval of source code, built to map natural language intent to real codebases.

It helps LLMs and developer tools find the right files and code snippets for a given user request.

What it’s for

• 🔍 Semantic code search
• 🤖 LLM / RAG context retrieval
• 🧠 AI copilots & agents
• 🧭 Navigating large or legacy repos

Installation

npm install inbed

Basic Usage

Create an embedder

import { OpenAIEmbedder } from 'inbed';

const embedder = new OpenAIEmbedder(
  process.env.OPENAI_API_KEY!
);

Initialize Inbed

import { Inbed } from 'inbed';

const inbed = new Inbed(embedder, {
  rootDir: process.cwd(),
  fileExtensions: ['.ts', '.js'],
  ignorePatterns: ['dist/**', 'node_modules/**']
});

await inbed.load();

Semantic Search

const results = await inbed.semanticSearch(
  'where cache is written to disk',
  5
);

results.forEach(r => {
  console.log(r.path, r.score);
});

Returns the most relevant files and snippets for the query.

Common Pattern: Prompt → Files

async function selectRelevantFiles(prompt: string) {
  const results = await inbed.semanticSearch(prompt, 5);
  return results.map(r => r.path);
}

Use this to feed only the necessary code into an LLM prompt.

How it works (short)

1. Scans the project
2. Splits files into chunks (AST-aware for TS)
3. Generates embeddings (cached locally)
4. Watches files for changes
5. Searches by vector similarity

Embedding Providers

• OpenAI
• Ollama (local)
• OpenRouter

What Inbed is not

• A static analyzer
• A refactor engine
• A bug detector

It retrieves relevant context — the LLM does the reasoning.