Getting Started

Introduction

Coreling is a local AI assistant that runs entirely on your machine. It auto-downloads everything it needs on first launch — no configuration, no API keys, no cloud.

How it works

Coreling is a Python TUI (terminal UI). On first run it silently downloads its inference engine into ~/.coreling/ and starts it as a background process. Your conversations, memory, and model weights never leave your machine.

Getting Started

Installation

One command. Coreling's inference engine and models are installed automatically on first run.

Prerequisites

  • Python 3.7+
  • 8 GB+ RAM recommended
  • macOS, Linux, or Windows

macOS / Linux

curl -fsSL https://coreling.org/install.sh | bash

Windows (PowerShell)

irm https://coreling.org/install.ps1 | iex

First launch

On first run Coreling downloads its inference engine (~50 MB) and an open language model (~2 GB). This only happens once. Subsequent launches are instant.

Getting Started

Quick Start

Start a session in one command.

coreling

You will see an engine selection menu:

  ▶ Uni-Core    Standard   Deep reasoning
    Multi-Core  Fast       Speed + math coprocessor

Use arrow keys to select, Enter to confirm. That is it — start chatting.

Example session

  ❯ you  What is the time complexity of quicksort?
  ● coreling  O(n log n) on average, O(n²) worst case.
              Use randomized pivot to keep it fast in practice.

  ❯ you  Remember: I work in Python
  ● coreling  Got it.

  ❯ you  /exit

Features

Engine Modes

Coreling offers two modes selectable at startup.

Uni-Core

Coreling's standard engine. Best for reasoning, writing, and coding. This is the default and recommended mode for most tasks.

Multi-Core

Coreling's fast engine. Routes arithmetic expressions through a deterministic coprocessor — evaluated locally without going through the language model, giving exact answers instead of approximations.

# Multi-Core detects math automatically:
  ❯ you  What is 1,847 × 293?
  ● coreling  541,171

Vision (automatic)

Drop any image path into the chat — Coreling detects it, ingests the image, and auto-switches to its vision model. No extra setup needed.

  ❯ you  /Users/me/screenshot.png What does this show?
  ● coreling  The image shows a Python traceback...

Features

Memory System

Memory is stored in ~/.coreling/brain.md — a plain Markdown file that acts as Coreling's persistent system prompt.

When you tell Coreling a fact, it silently appends a [Learned Memory] entry to brain.md. The memory is injected into every future session automatically.

# brain.md (example after a few sessions)
You are Coreling, an advanced AI orchestrator...

[Learned Memory]: User works at a fintech startup
[Learned Memory]: User prefers Python 3.12 one-liners

Edit it directly

brain.md is a plain text file. Open it in any editor to review, edit, or delete memories manually. Use /wipe to reset it to factory defaults.

Features

Image Analysis

Paste any image file path into your message. Coreling copies it to ~/.coreling/saved/, encodes it, and switches to its vision model automatically — no extra command needed.

  • Supports .png, .jpg, .jpeg
  • Works on macOS, Linux, and Windows paths
  • Vision model is pulled on first image use (~4 GB, one-time)
  • Session stays in vision mode for subsequent messages

Reference

Commands

Launch commands

coreling           # Start a new session
coreling --update  # Update to the latest version

In-session commands

/clear   # Reset conversation history (keeps brain.md)
/wipe    # Reset brain.md to factory defaults
/exit    # Quit Coreling

/clear vs /wipe

/clear resets the current conversation but keeps your learned memories. /wipe erases everything in brain.md — use it to start completely fresh.

Reference

Storage Layout

Everything Coreling needs lives in ~/.coreling/.

~/.coreling/
  brain.md        # system prompt + learned memories
  corelingd       # inference engine binary (auto-downloaded)
  artifacts/      # model weights (auto-downloaded)
  saved/          # images you've shared in chat

To reclaim disk space, delete ~/.coreling/artifacts/. Models will be re-downloaded on next launch. To fully uninstall, delete the entire ~/.coreling/ directory and the coreling binary.

Reference

Updating

Coreling checks for updates automatically in the background. When a new version is available you will see a notice at the end of your session.

# Update to the latest version
coreling --update

This re-runs the install script for your platform (macOS/Linux: bash, Windows: PowerShell). Your brain.md and saved images are not affected.