Skip to main content

AI Chat

Overview

Vipr Desktop's AI chat lets you discuss files, functions, metrics, and findings with an AI model. Every chat is grounded in the analysis context Vipr already computed: the file's source, its metrics, and its top findings. Answers are therefore specific to your code rather than generic advice.

AI Chat requires a Pro license and a configured AI provider.

Starting a chat

Use the Chat with AI action anywhere it appears:

  • File and function detail pages
  • Issues and anti-pattern views ("Validate Findings")
  • The source inspector drawer

Each launch opens the chat page with a new conversation scoped to what you were looking at.

Conversation history

Conversations are saved on your machine, in the same per-repository database as your analysis data:

  • The sidebar lists every conversation for the current repository, newest first.
  • Conversations get automatic short titles after the first exchange; rename or delete them from the sidebar.
  • Reopening a conversation restores the full thread, and you can keep asking follow-ups. This works after a restart, and via a vipr://navigate/ai-chat/<id> deep link.
  • Stopping a reply mid-stream keeps the partial text.

Nothing about your conversations is sent anywhere except to the AI provider you configured.

Providers

Configure a provider in Settings → AI:

  • Bring your own key. Anthropic (Claude), OpenAI (GPT), or Google (Gemini) API keys, stored encrypted with your operating system's keychain. Only current models with reliable structured output are offered.
  • Claude Code (subscription). If you have Claude Code installed and signed in, Vipr can run chats through your own local CLI so usage draws on your existing Claude subscription. Vipr never sees or stores your Anthropic credentials; subscription access for third-party apps is subject to Anthropic's policies and may change, and you can switch to an API key at any time.
  • AWS Bedrock. Use Claude models through your own AWS account (AWS profile, SSO, environment variables, or access keys). Credentials stay on your machine; short-lived tokens are minted per session.
  • Ollama. If a local Ollama server is running, your installed models appear automatically and run fully on-device.
  • Local model. A curated GGUF model downloaded and run by Vipr itself, fully on-device (see Local AI Setup).

Before anything is sent to a cloud provider, Vipr scans the outgoing context for strings that look like secrets and asks you to review them.

Documentation