Hiveryn
Work in progressHiveryn is a private, actively evolving system - no public link yet.
// 01. The problem
Coding agents are powerful but ungoverned. Run a few of them across a real codebase - or several codebases - and the gaps show fast: no shared state between sessions, no safety boundaries around what an agent may do, and no operational visibility into what is actually running. Hiveryn is my answer: an environment where agent sessions are first-class, coordinated, and accountable.
// 02. The system
Hiveryn is local-first and built independently, end to end: a Go daemon with a reusable agent runtime, an Electron/React desktop app, and shared domain components between them. It coordinates three kinds of agent sessions - architect, ticket and freeform - across multiple repositories at once. Work flows through filesystem-backed kanban boards; sessions get real PTYs, streaming status and events, and SQLite-backed state. Adapters plug in the agents themselves: Codex, Claude Code, and OpenCode.
// 03. Trust & safety
An environment that runs agents has to decide what they are allowed to do. Hiveryn gates consequential actions behind explicit approvals, exposes capabilities to agents through scoped MCP tools rather than open-ended access, and keeps every session scoped to the repositories and ticket it was created for. The goal is agents you can leave running - because the boundaries hold.
// 04. Operational reliability
Long-running agent sessions fail in ordinary ways - crashes, disconnects, restarts - so the system is built to survive them: crash recovery, resumable sessions, and a streaming event model that lets the desktop app reattach to whatever the daemon is doing. State lives in SQLite and on the filesystem, so nothing depends on a process staying alive.
// 05. Status
Hiveryn is shipped end to end and in daily use - it runs my own multi-repo work, including parts of this website. It is also very much in progress: the product is evolving weekly, and it is not ready for general use yet.