AI Assistant
The AI Assistant is xyva.ai's direct AI link — a conversational interface that understands your project, your tech stack, and your test results. Open it from the sidebar under AI Chat to start a session.
What the assistant Does
- Architectural advice — ask about component structure, dependency graphs, or migration strategies and get answers grounded in your actual codebase.
- Code review — paste a snippet or point to a file; the assistant highlights issues, suggests improvements, and explains trade-offs.
- Testing strategy — brainstorm coverage gaps, prioritize test types, or generate a test plan for an upcoming feature.
- Context-aware answers — the assistant automatically ingests your project's file tree, recent test results, and architecture vault data so every response is project-specific.
Supported Providers
| Provider | Model examples | Connection |
|---|---|---|
| Google Gemini | gemini-2.0-flash, gemini-2.5-pro | API key |
| OpenAI | gpt-4o, gpt-4.1 | API key |
| Anthropic Claude | claude-sonnet-4, claude-opus-4 | API key |
| Azure OpenAI | Any deployed model | Endpoint + key |
| Ollama | llama3, codellama, mistral | Local URL (no key) |
Configure your provider in Settings > AI Provider. The assistant validates the connection before the first message.
TIP
Running Ollama locally? Set the base URL to http://localhost:11434 and leave the API key empty. No data leaves your machine.
Persistent History
Every conversation is stored locally in the Electron user-data directory. Reopen a previous chat from the history sidebar to continue where you left off. History is never sent to any external service.
Core Components
The assistant is rendered by AiChat.tsx and powered by ai.service.ts on the Electron side. All provider routing, token counting, and streaming logic lives in the main process.
Sub-Pages
| Page | Content |
|---|---|
| Chat Interface | Message input, streaming, multi-turn conversations |
| Project Context | How the assistant knows your project |
| Code Generation | POM generation, test scaffolding, refactoring |
| System Instructions | Custom prompts, templates, per-project tuning |