Letronna — Your agent, with its own cloud and wallet

# Letronna > Letronna is an open-source AI agent, built in Rust, with its own cloud and wallet: it works on your computer and in your terminal with any model or account, hosts what it builds at a live link, and pays for the APIs it needs. macOS, Linux and Windows. Letronna is made by Gadlus LLC. The agent runtime is Apache-2.0 on GitHub; hosted models and cloud deploy are the paid parts. Contact: hi@letronna.ai ## Key facts - Runs locally on macOS 13+, Windows 11 and Linux (glibc 2.31+) - Memory and context are plain-text files under ~/.letronna - Works with any OpenAI-compatible endpoint, Anthropic, and local models via Ollama or llama.cpp - No account needed for local use; nothing leaves the machine unless a hosted model is connected - Capabilities: keeps session context, persistent memory on disk, local file and command access, scheduled runs, subagent delegation, MCP and shell tool integration ## What it does - Desktop app (macOS, Windows, Linux) and a terminal app that share the same sessions and memory - One app for many models: hosted models on a Letronna account, your own API keys (OpenAI, Anthropic, Gemini, xAI, OpenRouter), local models through Ollama, any OpenAI-compatible gateway of your own, and a ChatGPT plan's Codex models through a built-in plugin - Routes across connected accounts by tier and falls back to the next one when an account hits its limit - Letronna Cloud: runs a task on Letronna's servers from a prompt, a repository or a folder on your computer, and keeps web apps and bots (Discord, Telegram, schedulers) online around the clock - Tools for files, shell commands, web search and fetch, a browser, subagents ("crew"), schedules and MCP servers; less used tools are loaded on demand to keep each request small - Written in Rust as one native binary ## Benchmarks - SWE-bench Verified, the 10 tasks of OpenHuman's harness benchmark, graded by the official evaluator, on DeepSeek V4.1 Flash: Letronna solved 10 of 10 at 376k tokens and $0.008