v1.0.2 — critical install fix

Critical fix for installed systems. v1.0.0/v1.0.1 generated agent MCP configs that spawned bun <bundled source> — a command no installed machine can run (no bun, no repo on disk). Agents therefore could never call task_complete, and cards hung in working forever. The installed binary now serves the MCP sidecar itself (rvmp mcp-sidecar, wired automatically), and rvmp doctor performs a live sidecar handshake so a broken MCP path can never again fail silently.

If you installed 1.0.0/1.0.1: re-run curl -fsSL https://codegent.io/install | sh (or rvmp update), then restart the daemon.

v1.0.1 — rvmp

codegent is now rvmp. Same product, new name: binary rvmp, data dir ~/.rvmp, npm package rvmp. Fresh installs only (no migration — v1.0.0 was days old). Domain moves to rvmp.ai soon; install stays at the current URL until then.

codegent v1.0.0

codegent v1.0.0

A local, browser-based orchestrator for AI coding agents — real PTY terminals, a kanban board that routes attention, and a review flow with real diffs. Local-only by design: no accounts, no cloud, no telemetry.

Install

curl -fsSL https://codegent.io/install | sh

(or grab the tarball for your platform below: extracts to bin/codegent + share/web)

Highlights

• Board: queue → running → waiting-for-input → review → done; auto-start with worker limits.
• Universal agent tier: content-free state detection (process tree, OSC, screen manifests) for Claude Code, Codex, Gemini CLI, Goose, OpenCode, Aider and more — your terminal's content never leaves the terminal.
• Review flow: file-by-file diff with viewed-marks, queued line comments batched back to the agent, stale/conflict tracking, squash/merge/rebase, PR tracking via gh.
• CLI + service: codegent, codegent task add, codegent doctor, codegent service enable (launchd/systemd).
• Remote access = your own tunnel: see docs/expose-safely.md.

AGPL-3.0.