Koryphaios documentation
Koryphaios is a native desktop AI workspace that orchestrates a team of coding agents against your real project — planning, delegating, self-reviewing, and letting you undo anything. These docs explain how it works and where the sharp edges are.
Install the native app
Download the latest complete build for your computer. The installer detects your operating system and architecture; it does not clone the source tree or require Bun, Rust, or a package manager.
$ curl -fsSL https://koryphaios.com/install | sh$ curl.exe -fsSL https://koryphaios.com/install.ps1 | powershell.exe -NoProfile -ExecutionPolicy Bypass -Command -On macOS, the command downloads and opens the DMG. On Windows, it downloads and opens the setup executable. On Linux x64, it installs the AppImage at ~/.local/bin/koryphaios. Review the small POSIX installer or Windows installer before running it if you prefer.
What Koryphaios is
Most AI coding tools are a single model in a chat box. Koryphaios is an orchestrator: a manager agent (“Kory”) plans a task, delegates pieces to specialist worker agents running in isolated git worktrees, runs a read-only critic over their work, and synthesizes the result — all against your actual repository, on your machine.
It runs as a native desktop app (Tauri + a local Bun backend). Your code and app state stay local, while the context you choose is sent to whichever model provider or agent CLI you invoke. You bring your own models — API keys or the agent CLIs you already have installed (Claude Code, Codex, Grok, Cursor, Devin, Cline, Antigravity).
Start here
How Koryphaios works
The manager / worker / critic loop, git worktrees, time travel, and real-time streaming. Read →
Providers & CLI agents
API keys vs. subscription CLIs, the current provider registry, and how to connect them. Read →
CLI agent limitations
The honest caveats — what a CLI harness can and can't do, per provider. Read →
Notes
An Obsidian-style knowledge graph, native to Koryphaios and readable by your agents. Read →
Memory
Universal, project, and session memory plus rules and preferences. Read →
Share your models
Lend your providers to a friend — and the sandbox that confines a shared CLI. Read →
Design principles
- Local-first. The backend runs on your machine; your code stays yours.
- Bring your own models. Any provider, API key or CLI, mixed freely.
- Reversible. Every AI change is a ghost commit you can rewind.
- Honest about limits. Where a CLI or the sandbox can't guarantee something, the app and these docs say so.