HiveRail runs the CLI agents you already pay for — side by side, in parallel — each in its own isolated git worktree, all wired to one shared hub of tools. No vendor lock-in. Your keys. Your machine. Bought once.
Most AI coding tools lock you to one model, rent you access by the month, and run your code through their cloud. HiveRail is built the other way around — it orchestrates the agents and accounts you already have, and it stays out of the middle.
Everything is built around one idea: many agents working your codebase at once, safely, sharing the same tools, where you can see them.
Every agent works in its own git worktree on its own branch, cut automatically. Agents never trample each other — or your main branch. Review the diff, keep what wins.
Claude Code, Codex, Gemini, Kimi, OpenCode, Cursor and your own commands, tiled in one window. Status dots and attention alerts tell you exactly which pane needs you.
Every agent you spawn routes through a single hub — a browser, your memory, the Design Room, Linear and your own MCP servers. Wire a tool once; nothing to install or configure per agent.
Any agent can design — every page is measured against real accessibility checks before it certifies. Share it as a private link: your reviewer opens it, points at an element, even tries CSS live on the page — and their feedback lands back in your agent's pane as its next instruction. New versions reach every open page instantly.
Cockpit for the grid, the Board for your backlog, the Meter Room for usage and cost, the Merge Yard to review and keep the winner — the whole workflow, one keystroke away.
The part nobody else has: your agent's work becomes a private, certified page — and your reviewer's feedback becomes your agent's next instruction. Watch one round trip.
HiveRail meets your existing backlog and playbooks — turn a ticket into a running agent, and give every agent the same skills in a click.
Connect Linear and your issues appear as a live board in a side pane, right beside your agents. Drag a card onto an agent and it starts working the ticket — a build brief, a PRD, or a test-driven task — on its own rail, and hands back a branch for you to keep. Tokens are encrypted on your device and never leave it.
A built-in library of agent skills from skills.sh — search it, install with a drop, and bind a skill to any agent so your whole hive runs the same playbook. HiveRail also picks up the skills already in your Claude, Codex and Gemini installs, no re-import. Browse the library →
From open folder to a running agent in your first minute — the first launch spins up a live demo agent for you.
Run one command, open the link it prints, and point HiveRail at any folder. It detects your installed agent CLIs and cuts an isolated git rail for every pane it launches — no install to sit through.
One agent or a full grid — pick a model per pane, hand each a mission, and they share the same hub of tools while working their own rail.
Watch the diffs come in, side by side. Merge the winner back with one click. The rest never touched your branch.
Real screenshots from the app.




No subscription, no seat licenses, no metering anxiety. HiveRail runs on your machine with your own agent accounts — buy it once and it's yours.
npx hiverail — it opens in your browser, and installs as an app whenever you want. Runs on Windows and macOS (needs Node.js — most developers already have it). Your API keys and code never leave your machine.npx hiverail — and HiveRail opens in the browser you already have. Want it as a standalone app with its own window and dock icon? Click Install — that's a lightweight browser feature (a PWA), not a 200MB download. It only needs Node.js, which most developers already have.