wtf is miko?

Miko is a desktop application built for security testing, specifically for gathering information and probing targets during penetration tests and security audits. It is written in TypeScript with a Tauri and React based interface, so it runs as a cross platform desktop client rather than a web app. The tool centers on an AI agent that keeps memory of what it has found across a session, using something the project calls a blackboard mechanism. This lets the agent remember earlier discoveries, reasoning, and actions across multiple rounds of testing instead of starting fresh each time. It also has an exploration mode that keeps digging into a target automatically, and a focus system that locks onto a target and tracks unfinished leads so nothing gets left half checked. Users can switch between expanding sideways to cover more of an attack surface or drilling down into one specific weak point, and progress is shown visually as the exploration happens. A session that starts as a quick test can be turned into a full project with one click, keeping all findings and history intact. Miko ships with 23 built in skills covering information gathering, code auditing, vulnerability probing, and reverse engineering, plus eight preinstalled tools including subfinder, katana, fscan, gobuster, and radare2. It supports multiple AI models, including DeepSeek, Qwen, and Azure OpenAI. For non-technical readers: this is not a general purpose coding assistant. It is a specialized agent meant for people doing authorized penetration testing or security audits, and its automated exploration mode can run real probing actions using tools like Bash and web requests. The maintainers warn against combining fully automatic permissions with exploration mode in untrusted environments, since sub-agents may wait on permission checks and time out. To get started, users download a portable Windows build from the project's Releases page, unzip it to a writable folder, and add an API key for their chosen AI model in settings. Building from source requires Bun and Rust, plus Visual Studio on Windows for the C++ desktop workload. The project is maintained by OpenAisec and welcomes issues and pull requests.