Skip to main content
Luminy is a local-first agentic AI coding assistant built as a native desktop app. Think of it like Claude Code or GitHub Copilot Workspace — but provider-agnostic and entirely under your control. You can wire it up to Anthropic, OpenAI, Gemini, or any of 20+ supported cloud providers, or you can run it 100% offline using a local model through Ollama. In every case, the AI is given direct, sandboxed access to your codebase and developer tools, so it can do real work rather than simply suggest code for you to paste.

​
How Luminy works

Luminy operates through a continuous agentic loop that keeps the model working until your task is complete.
  1. You send a message — a question, a task description, or a request to fix a bug.
  2. The LLM reasons and selects tools — the model decides what information it needs or what actions to take, then emits one or more tool calls (e.g. read a file, run a shell command, search for a symbol).
  3. Tools execute locally — Luminy runs each tool directly on your machine: reading and writing files, executing shell commands, querying git history, or reaching out to an MCP server.
  4. Results feed back into context — tool outputs are returned to the model, which reasons over them and either calls more tools or produces a response.
  5. Repeat until done — the loop continues, compacting context automatically when conversations grow long, until the task is finished or you intervene.
This loop means Luminy can handle multi-step tasks — refactoring a module, writing and running tests, diagnosing a failing build — without you having to copy-paste anything.

​
Key capabilities

20+ AI Providers

Connect to Anthropic, OpenAI, Gemini, GitHub Copilot, OpenRouter, xAI, Mistral, Groq, and many more — or run fully offline with local Ollama models.

Code Intelligence

Tree-sitter indexing gives the AI a structural understanding of your code across 8 languages: Python, JavaScript, TypeScript, Rust, Go, Java, C++, and C#.

Agentic Tools

The AI can read, write, and edit files; execute shell commands; interact with git; search your codebase by symbol or content; and browse the web.

MCP Integration

Model Context Protocol support lets you connect any external data source, API, or service as a first-class tool the AI can call during its reasoning loop.

​
Local-first & private

Every piece of data Luminy creates — your sessions, chat history, project metadata, and settings — is stored entirely on your own machine. Luminy has no account system and no cloud sync. The only network requests Luminy makes are the ones you explicitly configure: calls to your chosen AI provider’s API, MCP server connections you define, and periodic checks for app updates. This architecture means you can use Luminy with sensitive or proprietary codebases without any data leaving your control. Pair it with a local Ollama model and the entire system runs air-gapped, with zero external dependencies at inference time.