Getting Started
From install to your first prompt in a few minutes.BYOK: model tasks need at least one configured model (a Gemini, Claude or OpenAI key, a local Ollama model, GitHub Copilot, or the Claude Code CLI signed in with your account). Slash commands and workspace tools work without any.
1. Open iNoU and accept the agreement
Click the iNoU icon in the Activity Bar and accept the short iNoU agreement. Nothing runs before that.
2. Configure your LLM models
Open Config ▸ LLM models. Every provider shows only the models it really lists: nothing is hardcoded.
- Google Gemini, Anthropic Claude, OpenAI: add the API key. Keys are verified with their provider before they are saved, and stored in VS Code SecretStorage. Organization-wide Anthropic keys also need the Workspace ID (
wrkspc_…). - Ollama: set the server URL; every model installed on the server is offered.
- Claude Code (CLI): iNoU finds the CLI or installs it with Anthropic's official installer, then you Sign in with your Claude account.
- GitHub Copilot: enable
inou.enableCopilotDetectionin the settings.
A provider that lists no models (server down, key refused) is not usable; the LOG tab says why.
3. Create your Developer Profiles
Open Config ▸ Preferences ▸ Developer Profiles. A Default profile is created with your first provider. Each profile has:
- a Plan Mode model: understands the prompt, answers questions and plans the change;
- an Act Mode model: writes the code;
- the prompt quality meter and iNoU interference switches.
Plan and Act can come from any providers, e.g. Gemini to plan and Ollama to act, or the Claude Code CLI with sonnet to plan and haiku to act. Pick the profile of each chat in the combo above the prompt.
4. Send your first prompt
- Type your instruction. Use
@to mention files, check files in Files, or attach images and documents. - Watch the Prompt Quality Meter: a vague task gets one question instead of a guess.
READ-MODEblocks every file change and command; switch toWRITE-MODEwhen you want iNoU to act.- Review the proposed changes and apply them.
5. Work without a model
Commands starting with / run locally, with no internet and no LLM credits: /? lists them, e.g. /find, /get diagnostics, /file read, /project run test, /trello. If VS Code closes mid-answer, /recover picks the task up.