Short answer
Happy, which started as Happy Coder, and Orchestrator.inc (formerly Agent Orchestrator) are both free, open source apps with desktop builds and iOS and Android apps. Happy is built around using your agents privately from any device: its sync is end-to-end encrypted, its relay can be self-hosted, and you can share a live session. Orchestrator is built around supervising many agents and following each pull request through CI and review. Choose Happy if end-to-end encryption between your phone and your machine is the requirement. Choose Orchestrator if you run several agents on real repositories and want CI and review feedback to reach each one without you relaying it.
At a glance
| Orchestrator.inc | Happy | |
|---|---|---|
| Best for | Supervising many agents through PR, CI and review | Using Claude, Codex and Grok from any device, privately |
| License and price | Apache-2.0, free locally; optional Cloud plans from $20/month | MIT, free; voice has a free tier and an optional subscription |
| Desktop | macOS, Windows x64, Linux | macOS, Windows, Linux (Happy Desktop) |
| Mobile | iOS and Android app with push notifications | iOS and Android app, plus a web app |
| Away from your network | Managed HTTPS tunnel; no built-in device relay yet | Happy's end-to-end encrypted relay, which you can also host yourself |
| Agents | 38 harnesses, each the vendor's own CLI | Claude, Codex and Grok models through Happy's own agent runtime; the original CLI wraps Claude Code |
| Isolation | Git worktree and branch per session; multi-repo workspaces | Git worktree on its own branch when two pieces of work share a repository |
| Overview | Kanban board placed by session, PR, CI and review state | Session list |
| After the push | Failing checks, review comments and merge conflicts sent to the owning agent | Not documented |
| Collaboration | One user; phones and other desktops connect to the same machine | Invite someone into a session; both can steer |
| Planner agent | Project orchestrator that briefs and spawns workers | Subagents inside a session |
The problem in practice
You start three agents before leaving for the day and check in from your phone on the train. One is waiting for a permission prompt, one has opened a pull request that fails a test, and one has a review comment from a teammate. Answering the prompt from your phone is easy in either app. The rest means noticing the red check and the comment and passing each one to the right agent, which is exactly the part a phone is worst at.
How each tool handles it
Task: get a worker's PR from a failing test to merged.
In Happy
- Start a session from the desktop or phone and work with the agent until the change is ready.
- Push the branch and open the PR on GitHub. Happy's docs do not describe tracking pull requests or CI.
- Watch the checks yourself. When one fails, open the session and paste the failure in, then do the same for review comments.
- Merge on GitHub.
In Orchestrator.inc
- Spawn a worker from the board, the phone app or with
ao spawn. Once it pushes, Orchestrator discovers the PR. - The local daemon sees the failing check and sends its name, link and the end of the failed log to the worker that owns the PR.
- The worker pushes a fix. Review comments go to the same worker, and so do a reviewer agent's findings if you set one up.
- The card moves to Ready and your phone gets a push. You read the diff and merge; Orchestrator never merges on its own.
Where Happy is a good fit
- End-to-end encryption is a requirement. Happy's relay forwards encrypted content it cannot read, and you can run the relay yourself.
- You want to share a session. Invite a colleague into a live session and both of you can steer the agent.
- You switch models often. Move between Claude, Codex and Grok models inside one conversation and keep the history.
Where Orchestrator fits better
- Workflows where CI and review feedback should reach the agent on its own. GitHub and GitLab, including self-managed GitLab, are polled with no webhooks to set up, and failing checks, requested changes and merge conflicts go to the owning session, de-duplicated.
- People who want the vendor CLIs themselves. Orchestrator launches Claude Code, Codex, Cursor, Gemini CLI and the rest of its 38 harnesses as they ship, so their own tools, settings and permission modes apply.
- People who want a second model to review the first. 22 of the 38 harnesses can act as reviewer agents, so Codex can review a Claude Code PR.
- Larger plans. A project orchestrator agent breaks a plan into tasks, spawns and briefs workers, and follows their progress across branches and PRs.
- People who run many sessions. Cards move across the board on their own, based on what is true of each session and its PR. Nothing is dragged by hand.
Orchestrator's phone app reaches your desktop over a managed HTTPS tunnel, and there is no built-in device relay yet, so Happy is ahead on private remote access. Orchestrator Cloud runs sessions in hosted sandboxes if you would rather not keep a machine on.
Pricing
Orchestrator is free and open source (Apache-2.0); you bring your own agent subscriptions. It uses the agent CLIs and sign-ins you already have, so usage is billed by your provider. Optional Orchestrator Cloud plans, for agents that keep working while your laptop is closed, are Starter at $20/month and Pro at $40/month, with Enterprise pricing on request. As of October 2026, Happy is free under the MIT license; only its voice feature has an optional subscription.
Who should pick which
Pick Happy if
- end-to-end encryption between your phone and your machine is a requirement
- you want to self-host the relay or share a live session with someone
- you mostly drive one or two agents and spend a lot of time on your phone
Pick Orchestrator if
- you want CI failures, review comments and conflicts sent to the owning agent without forwarding them
- you want the vendor CLIs themselves, or an agent beyond Claude, Codex and Grok
- you want a planning agent that spawns workers
- you run many agents and want a board with live PR and CI state
Try it
Install Orchestrator, add a repository, and spawn your first two workers with the quickstart; then pair the phone app and let a PR fail CI to see where the failure lands.
Frequently Asked Questions
Is Happy Coder the same as Happy?
Yes. Happy Coder was the original name of the CLI and phone app for continuing a Claude Code session from your phone. The project is now called Happy and adds a desktop app with its own multi-model runtime. The original CLI is still available as happy on npm.
Is Happy open source and free?
Yes. As of October 2026 Happy is MIT licensed and free, and only voice has an optional subscription. Orchestrator.inc is Apache-2.0 and free to run on your own machines.
Is Happy end-to-end encrypted?
Happy says so: content is encrypted on one device and decrypted on another, and the relay only forwards it. Orchestrator.inc's same-network connection stays on your network. Away from it, Orchestrator uses a managed HTTPS tunnel, and it has no built-in device relay yet.
Does Happy handle pull requests and CI?
Happy's docs do not describe pull request, CI or review tracking as of October 2026. Orchestrator.inc follows each worker's PR and sends failing checks and review comments back to it.
Can I use Claude Code from my phone with Orchestrator.inc?
Yes. The Orchestrator.inc phone app shows each session's chat or terminal, lets you answer approvals and start workers, and sends a push when a session needs you or a PR is ready to merge. You can download Orchestrator.inc from orchestrator.inc and pair the phone app in a few minutes.