Every one of these four will run a coding agent for you. The interesting question isn’t can they run code — it’s where can you run them from. I’m looking at CodeNomad, Kandev, Paseo, and Orca through a mobile-first lens, because that’s the axis they actually separate on. Two of them ship real phone apps. One is a desktop cockpit that happens to have a browser mode. One is a governance workbench that assumes you’re sitting at the desk. That difference tells you more about each project’s philosophy than any feature list.
I’ve run or deployed three of the four on my own boxes, so this is written from the seat of someone who actually uses agent orchestration day to day — a Go binary on a self-hosted VM, a fleet of agents on worktrees, and a phone that wants in on the action.
CodeNomad: the OpenCode-only cockpit
2,589★ · TypeScript · MIT · created Nov 2025 · pushed Sep 24, 2026
CodeNomad takes OpenCode — a terminal-native agent — and gives it a premium desktop workspace. Electron and Tauri builds, multi-instance sessions side by side, voice input, sidecars for VSCode and terminal tabs, command palette, theming, git worktrees, session management. It does exactly one thing, and it does it well: make OpenCode sessions visible and controllable.
- Desktop apps: macOS (Universal), Windows, Linux (Tauri deb)
- Server mode:
npx @neuralnomads/codenomad --password <pw> --launch— a password-protected browser UI you can reach from anywhere - Single-provider by design: this is the cockpit for OpenCode, not a multi-agent platform
Mobile verdict: browser-only. The server mode technically opens on a phone browser, but nobody is navigating a worktree dashboard on a 6-inch screen by choice. There is no iOS or Android app, and the product’s energy is squarely on desktop polish. If you live in OpenCode and want a GUI on your dev machine, this is the right answer — I ran it on my dev box, and it’s genuinely pretty. It’s just not a tool that travels.
Kandev: the governance workbench
837★ · Go · AGPL-3.0 · created Jan 2026 · pushed Sep 24, 2026
Kandev is the open workbench for coding agents: plan → review → PR, with human approval gates at every step the operationally sensitive way. It’s agent-neutral through ACP (Agent Client Protocol) and terminal-native CLIs — Claude Code, GitHub Copilot, Gemini CLI, OpenCode, Cursor — with a Go orchestrator coordinating tasks and an agentctl sidecar translating Kandev’s protocol into whatever the agent actually speaks. Executors span local, worktree, Docker, Sprites ephemeral cloud, and SSH remote hosts.
- Review-first philosophy: humans review every change before it merges
- Kandev MCP lets a coordinator agent spawn subtasks, attach branches, hand work off
- Integrations run deep: GitLab, Jira, Linear, Sentry, Slack
- Self-hosted single binary — this is the one I’ve got running on a dev VM behind HAProxy (v0.83.0)
Mobile verdict: present-but-desktop. The web UI renders in any phone browser, but that’s not the point of Kandev. The whole design assumes a human at the review desk: approving plans, commenting on diff steps, deciding what ships. That’s a feature, not a bug — it’s exactly why I trust it for real work. But “check on a run from the train” isn’t a flow Kandev builds for. It’s the tool you ship work with, and the price is that it wants you present.
Paseo: the control plane with a pocket engineer
18,440★ · TypeScript · custom license · created Oct 2025 · pushed Sep 24, 2026
Paseo’s pitch is one line: run any coding agent from anywhere. Claude Code, Codex, Copilot, OpenCode — orchestrated from a self-hosted control plane with desktop clients and a native mobile app. The headline here is the iOS app, “Paseo Pocket Engineer”, which is exactly the thing the first two tools don’t have: a purpose-built phone surface for steering agents that are running elsewhere.
- Desktop app for Mac, iOS app for the phone
- Multi-provider: doesn’t care which agent you run underneath
- Self-hosted, open-source-adjacent — read the license before you build on it. It’s a custom source-available license (one of the few “OSI-ish” asterisks in this list)
Mobile verdict: the first bridge. Paseo is the first of the four to take “run agents from your phone” seriously as a first-class product surface, not a browser fallback. It’s the established, actively-shaped option if your baseline requirement is “phone in my pocket can reach my fleet.” The license and the proprietary lean are the things to sanity-check before a company bet, but as a tool for a self-hosted operator it earns its 18.4k stars.
Orca: the parallel agent IDE with real companions
77,459★ · TypeScript · MIT · created Mar 2026 · pushed Sep 24, 2026
Orca is the rocket of the group: 77k stars in six months, YC-backed, and it looks like it. The agent development environment runs Claude Code, Codex, and anything else in parallel, each in its own worktree, with terminals, diffs, an embedded browser, and a CLI in one app. Then it goes where none of the others went: native iOS and Android companions (App Store “Orca IDE”; Android APK from the stablyai/orca releases), plus a remote runtime.
- Parallel agent fleet, per-worktree isolation, review AI output in-app
- Desktop: macOS, Windows, Linux
- Mobile: iOS app + Android APK — run a fleet, validate promo codes, update e2e snapshots, review tax-logic diffs from the phone
- MIT license, and it means it
Mobile verdict: the one built for the phone. Orca isn’t “accessible from mobile” — its mobile companion is part of the advertised core experience (“Work on-the-go”). That’s the strongest mobile story of the four, and it’s backed by the most aggressive velocity. At 77k★ and six months old, expect rough edges and fast breakage — but this is the tool to actually try, not just watch.
The verdict
Ranked for the mobile-first operator:
Orca — The mobile-native pick. Parallel agents in isolated worktrees, real iOS + Android companions, MIT. The fastest-moving project in the space for a reason. Rough edges at this velocity are the tradeoff.
Paseo — The “from anywhere” pick. First true desktop→mobile bridge for steering a fleet, and the established option with 18.4k★. Verify what its custom license means for your infra.
Kandev — The “ship work with it today” pick. Governance, approval gates, and review-first design make it the trustworthy workbench — and make its mobile story intentionally weak. Present-by-design.
CodeNomad — The “I live in OpenCode and want a GUI” pick. Gorgeous desktop cockpit, single-provider, browser-only for remote. If you run multiple agents or want phone reach, it’s the wrong shape.
The pattern is clear: the projects that treat mobile as a first-class surface — Orca, Paseo — are the ones betting on agent fleets you manage, not sessions you supervise at a desk. The workbench approach (Kandev, CodeNomad) assumes the developer is the bottleneck and wants them physically present. Both philosophies are legitimate; they just answer different questions. If your question is “how do I keep an eye on my agents without carrying a laptop,” the market has picked its answer, and it ships with a phone app.