Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.khal.ai/llms.txt

Use this file to discover all available pages before exploring further.

​
Prerequisites

  • Bun v1.1+ or Node.js 20+
  • PostgreSQL 15+ with the pgvector extension
  • A Gemini API key for embeddings (GEMINI_API_KEY)

​
Install

bun add @khal-os/brain
The package includes both a TypeScript library and a brain CLI.

​
Create Your First Brain

1

Initialize a brain vault

brain init my-brain
This creates a my-brain/ directory with:
  • brain.json — configuration file
  • memory/ — default memory directory
  • Ready-to-use Obsidian vault structure
2

Set your Gemini API key

Brain uses Gemini Embedding 2 for multimodal vector embeddings.
export GEMINI_API_KEY="your-key-here"
Also accepted: GOOGLE_API_KEY. Either works.
3

Add content to your vault

Add markdown files to your brain directory. Brain understands standard markdown with optional YAML frontmatter:
---
title: "Architecture Decision: Use PostgreSQL"
type: decision
confidence: high
---

We chose PostgreSQL over SQLite because we need
full-text search (tsvector), vector similarity
(pgvector), and concurrent access from multiple agents.
You can also mount external directories:
brain mount ./docs --brain-id my-brain
4

Index and embed

Scan files, chunk them, and generate embeddings:
brain update --brain-id my-brain --embed

[brain] Scanning my-brain...
[brain] Chunking: 12 documents → 47 chunks
[brain] Embedding: 47/47 chunks
[brain] Done (12 files, 47 chunks, 47 embedded, 3.2s)
5

Search

brain search "why did we choose postgres" --brain-id my-brain

── Results (3 hits, strategy: rag) ──────────────
1. [0.92] Architecture Decision: Use PostgreSQL
   We chose PostgreSQL over SQLite because we need
   full-text search (tsvector), vector similarity...

2. [0.71] Infrastructure Setup Guide
   PostgreSQL 15 is required. Install pgvector...

3. [0.65] Database Migration Notes
   Migration 001 creates the brains table...
Each result includes a confidence score (0-1). Brain drops low-confidence noise by default.

​
Using the TypeScript API

import { initBrain, updateBrain, searchBrain } from '@khal-os/brain'

// Create a brain
const brain = await initBrain({
  name: 'my-brain',
  path: './brain',
  brainType: 'engineering',
})

// Index files and generate embeddings
await updateBrain({
  brainId: brain.id,
  brainPath: './brain',
  embed: true,
})

// Search
const results = await searchBrain('why did we choose postgres', {
  brainId: brain.id,
  limit: 5,
  minConfidence: 0.5,
})

for (const hit of results.hits) {
  console.log(`[${hit.confidence.toFixed(2)}] ${hit.title}`)
  console.log(`  ${hit.snippet}`)
}

​
CLI Commands

CommandDescription
brain init <name>Create a new brain vault
brain update --brain-id <id> --embedIndex files and generate embeddings
brain search "<query>" --brain-id <id>Search with hybrid BM25 + vector
brain health --brain-id <id>Check vault health and quality
brain listList all registered brains
brain mount <path> --brain-id <id>Mount an external directory
brain analyze --brain-id <id>Analyze brain structure and coverage
brain grade --brain-id <id>Grade brain quality (A-F)

​
Next Steps

Search Deep Dive

Learn about BM25, vector, and hybrid strategies, confidence scoring, and strategy segments.

Configuration Reference

Full brain.json reference, environment variables, and embedding model options.