Show me the screen.
I’ll tell you what broke.

Chelp AI reads errors, dialogs, terminal output, and code — then explains the problem and the fix in plain language. Cloud models when you want them, local models when you don’t.

v1.0.0 Windows x64 No admin required

Screenshots happen only when you click. Nothing leaves your PC until you press Send.

Chelp AI — Dashboard
The Chelp AI desktop app showing the dashboard, screen analysis, and a loaded local model

Bring your own model

DeepSeek OpenRouter OpenAI Gemini Ollama LM Studio
Three steps

From “what is this?” to fixed.

No prompt engineering, no copy-pasting screenshots into a chat box. Show, ask, fix.

01

Capture

Pick a display, a window, or draw a region. Preview the exact image and brush over passwords before anything is sent.

02

Ask

“Why did this crash?” or nothing at all — Chelp reads the screen and starts from what is actually visible.

03

Fix

Get the problem, the likely cause, and numbered steps. Facts and guesses are marked separately.

Features

Small app. Sharp answers.

Everything is built around one idea: the screen is the context.

Screen Analysis

Displays, windows, or a drawn region on any monitor. Preview and redact sensitive spots before sending.

Guard Mode

A permanent scope lock. Jailbreaks and harmful requests are refused locally, before a single byte reaches a provider.

Local Models

Chelp can install Ollama itself, pull models with live progress, and keep the active model warm for two hours.

Floating Assistant

A compact always-on-top window that collapses to an orb. Ask without leaving the app you are debugging.

Your Model, Exactly

The answer always comes from the profile marked Active. No silent swaps, no surprise Qwen. Backup AI is opt-in and labeled.

Private by Default

Keys are encrypted with Windows DPAPI. Screenshots stay on disk until you send. History is local and can be wiped.

Always on

A helper. Not a Swiss Army knife.

Chelp is built to explain what broke — not to run commands, browse for you, or become someone’s tool. Guard Mode sits in the main process, before any network request.

  • Only errors, troubleshooting, software, settings, and general questions.
  • Jailbreaks and harmful requests are blocked on-device.
  • Text inside screenshots is treated as data, never as instructions.
  • There is no off switch in the UI — and no tool access to abuse.
Guard Mode active

Allowed

  • ✓ Diagnose a crash or error
  • ✓ Explain a setting or command
  • ✓ Review code shown on screen
  • ✓ Answer general questions

Blocked locally

  • × Malware and credential theft
  • × Bypassing security or licensing
  • × Prompt injection from a screenshot
  • × “Ignore your rules” requests
0 bytes sent on a blocked request
Download

Take it for a spin.

One app, two ways to run it. The site always points at the latest GitHub release — with a direct build as a fallback.

Portable

For USB drives and locked-down PCs
.zip

A zipped app folder you can unpack anywhere. No installer, no registry entries, no Start menu icon.

  • Unzip and run from any folder
  • Settings still live in your user profile
  • Same app, same models, same Guard Mode
Download Portable 120 MB Chelp-AI.1.0.0-portable.zip
Direct download · v1.0.0 Windows 10/11 · x64 · ~500 MB free All releases on GitHub
FAQ

Before you ask.

Do I need an API key?

For cloud providers, yes — paste a key or use one-click OpenRouter sign-in. Local models through Ollama or LM Studio need no key at all.

Does it take screenshots on its own?

No. A capture happens only when you click Capture or press the hotkey. You see a preview, can redact parts of it, and nothing is uploaded until you press Send.

Can I use it completely offline?

Yes. Chelp can install its own Ollama runtime and download a model, then work with no network connection. Cloud profiles are optional.

Can I turn Guard Mode off?

No. It is part of the request pipeline, not a setting. The model is also told it cannot run commands or control your computer — and the app has no tools to do it.

Which models are supported?

DeepSeek, OpenRouter, OpenAI, Google Gemini, any OpenAI-compatible endpoint, and local models via Ollama or LM Studio. Vision models are marked, so screen analysis uses a model that can actually see images.

Is there a macOS or Linux build?

Not yet. Chelp AI is Windows 10/11 (x64) only for now — it leans on Windows capture, the tray, DPAPI, and native window behavior.

How do updates work?

Download the latest release and run it over the old version. Conversations and settings stay in %APPDATA%\Chelp AI.