Koderithm (VS Code Extension)

Minimal coding assistant that can run fully offline at user level (no admin privileges required for extension usage).

What this MVP includes

One-click ON/OFF toggle from status bar (Ctrl+Alt+A).
Offline provider by default (local endpoint, no login required).
Optional online provider (explicitly disabled unless enabled in settings).
Prompt -> code generation in active editor (Ctrl+Alt+G).
Run + test command execution (Ctrl+Alt+R).
Background auto-fix after failures (Ctrl+Alt+F manually, or automatic on test failure).
Profile switch:
- Lab: simpler student-style code.
- Project: more structured and robust code.

Smart Model Selection (v0.0.4)

The extension now keeps two models on the device (lightweight + best‑for‑your‑hardware). If an installation ever fails or stops, it will retry automatically until both are present.

The extension automatically detects your device specifications on first activation and installs the best-suited models:

Lightweight Model (qwen2.5-coder:1.5b): Installed immediately - fast, low resource usage
Best Model: Auto-selected based on your device specs and downloaded in the background

Device Tiers

Tier	Requirements	Best Model	Size
Very‑Low	< 4GB RAM or < 2 cores	qwen2.5-coder:1.5b	1GB
Low	4‑7GB RAM, 2+ cores	qwen2.5-coder:3b	2GB
Mid	8‑11GB RAM, 4+ cores	qwen2.5-coder:7b	5GB
High	12‑15GB RAM, 6+ cores	deepseek-coder:13b	8GB
Very‑High	16‑31GB RAM, 8+ cores	deepseek-coder:33b	20GB
Ultra	32GB+ RAM, 12+ cores	deepseek-coder:65b	40GB

Using Different Models

You can switch between models anytime:

Koderithm: Use Lightweight Model (Fast) - Immediate response, lower quality (always available)
Koderithm: Use Best Model (Quality) - Better results, available once downloaded (background process)

The extension intelligently falls back to lightweight if the best model fails.

User-level setup (no admin)

The extension will automatically attempt to install Ollama and pull the default model on first activation. If automatic setup fails, follow manual steps below.

Manual setup (if needed)

Install Node.js in user scope (portable or user installer).
In this folder:
```
npm install
npm run compile
```
Press F5 in VS Code to launch Extension Development Host.

To package/install locally without admin:

npx @vscode/vsce package --allow-missing-repository
& "$env:LOCALAPPDATA\Programs\Microsoft VS Code\bin\code.cmd" --install-extension .\koderithm-0.0.1.vsix

Offline model setup

Default endpoint is http://127.0.0.1:11434/api/generate (Ollama-compatible). Default model is qwen2.5-coder:1.5b.

The extension will automatically pull the model on first use. If needed manually:

ollama pull qwen2.5-coder:1.5b

You can also point to any local OpenAI-compatible endpoint by changing:

deviceAssistant.offline.endpoint
deviceAssistant.offline.model

Keep deviceAssistant.enableOnline = false to stay fully offline.

Commands

Koderithm: Toggle On/Off
Koderithm: Generate Code From Prompt
Koderithm: Run and Test Project
Koderithm: Auto-Fix Last Failure
Koderithm: Run Doctor (Environment Check)
Koderithm: Switch to Lab Profile
Koderithm: Switch to Project Profile
Koderithm: Switch to Offline Provider
Koderithm: Switch to Online Provider

Troubleshooting

If commands appear to do nothing:

Run Koderithm: Run Doctor (Environment Check) (Ctrl+Alt+D).
Make sure your local model endpoint is reachable (default: http://127.0.0.1:11434/api/generate).
Set deviceAssistant.runTestCommand if your project has no detectable test command.
Open Output panel and select Koderithm to see command logs/errors.

Notes

All shell commands run in the current user context through the VS Code extension host.
No administrator elevation is requested by this extension.
Use responsibly and comply with your institution/company policies.

Koderithm

KodeRithm