Introduction

Welcome to the official developer documentation for Kyvora Studio, the next-generation AI-native IDE built from the ground up for professional developers. Conventional editors treat your source files as individual, disconnected text buffers. Kyvora is different: it compiles and continuously indexes your entire workspace into a semantic graph, enabling autonomous agent operations, real-time multi-file completions, and zero-breakage refactoring cycles.

SEO & Engine Optimization: Kyvora is built to support modern full-stack web frameworks (Next.js, Remix, Astro), backend systems (Spring Boot, Go, Rust), and complex microservices. This documentation explains the syntax graphs, agents, and customization files that power the workspace.

Kyvora Overview & Vision

Software engineering has evolved past basic code-completion wrappers. Kyvora's goal is to transition developers from writing line-by-line syntax to directing autonomous agents. By combining static code analysis with deep LLM orchestration, Kyvora understands:

How type changes in your backend database schemas propagate to your frontend React properties.
Your codebase's architectural patterns, import guidelines, and testing frameworks.
How to build, compile, and run your project's background test runners to verify changes before writing them to disk.

Codebase Context Graph

At the core of Kyvora is the Context Graph. As soon as you open a folder, a background daemon parses your files, constructing an AST (Abstract Syntax Tree) representation of all declared classes, variables, route mappings, and exports. This graph continuously watches your filesystem. When you trigger the AI agent, it doesn't just read the open file—it traces import paths, reviews database tables, and analyzes matching types to ensure a complete understanding of your repository state.

Core Capabilities

Multi-File Autocomplete

Traces structural dependencies across files to suggest accurate type completions and import blocks as you type.

Autonomous Action Loops

Spawns background compiler agents that execute, compile, run test suites, and fix diagnostics autonomously.

Offline LLM Runtime

Support for local endpoints (Ollama, LLaMA, LM Studio) to execute inference tasks offline for data privacy.

Smart Error Parsing

Analyzes build terminal output stack traces to locate source bugs and propose one-click hotfix patches.

Architectural Design

Kyvora utilizes a dual-engine architecture designed for speed and reliability. The local client handles keybindings, file systems, and LSP indexing, while a separate sandboxed daemon orchestrates background tasks, terminal command execution, and AI API interfaces. This separation guarantees that your editor remains responsive even during heavy LLM token streaming or complex background test runs.

Performance & FAQ

Q: How does Kyvora handle large repositories?
A: Kyvora splits indexing into chunks, using system indexers to exclude build output files (like `.next/`, `node_modules/`, `target/`). Memory consumption stays low by unloading inactive AST blocks.

Q: What AI models does Kyvora support?
A: Kyvora integrates with OpenAI, Anthropic Claude, Gemini, DeepSeek, and local endpoints via Ollama or LM Studio.