Published Updated
Questions people actually ask.
If yours is not here, the docs go deeper, and the glossary defines every term on this site.
-
How is Orbit licensed?
Orbit is source-available (Business Source License 1.1; each version becomes Apache 2.0 after four years).
-
What is Orbit?
A desktop app that gives coding agents somewhere to work. You post work to one durable queue, every run gets its own copy of the code, an inspector checks the result against git, and a permanent ledger records what shipped. The agents are the ones you already use — Orbit is the office around them.
-
Is Orbit only for running agents?
No. It covers the loop around the work as well. Sivants Studio makes the assets — images, icons, video, skyboxes, 3D worlds. Sweep checks that the interface you built matches the mockup it came from. The agents do the coding. The daily summary and the completion ledger keep the history. Dispatch builds and releases to TestFlight, the Play Store, Vercel or Railway and reports the deployment status back. Git integration holds the branches, the worktrees and the verdicts together. Security mode scans the project, lays out what it finds, and puts an approved fix through the same verification as any other change.
-
How is this different from running agents in terminal tabs myself?
Tabs do not survive a reboot, do not hold a queue, and cannot tell you what happened while you were away. Orbit keeps one queue across every project and every machine, gives each run a private copy of the code so nobody works on top of anyone else, and reads its verdicts out of git rather than from the agent. You can queue work before bed and read a ledger in the morning.
-
Which agents can I use?
Claude Code, Codex, Gemini, Grok, DeepSeek, Kimi, GLM, Llama, anything OpenRouter fronts, and anything else OpenCode can run, under whichever harness you prefer. They share the same queue, the same context and the same vault, so swapping the model does not change the workflow.
-
Can I use my Claude or ChatGPT subscription instead of buying API tokens?
Yes, and that is the default. Every model tier ships set to subscription mode, so Orbit drives the agent through the login its CLI already has, and the work counts against that plan rather than billing per token. Bringing your own provider key works too — Anthropic, OpenAI or Google — and the cost display disappears when you do, because what the agent spends is then between you and that provider. Sivants Credits are there if you would rather manage neither, with the per-message cost shown before you send.
-
What do I need to run it?
Windows, macOS, or Ubuntu (Linux). The desktop app installs on Windows 10 or later and on Apple Silicon Macs; Ubuntu 22.04 and 24.04 projects run through the Orbit Linux helper, with the window on Windows. You also need a coding agent on your PATH — Claude Code, Codex CLI or Antigravity — which Orbit can detect and configure during onboarding. A provider key is optional.
-
Do agents trip over each other in the same repository?
No. Every run gets its own git worktree, so two agents working the same repository never share a working copy and never have to be told to wait their turn. Memtrace ships with Orbit and indexes the code underneath all of it: ask it what a change would break before anyone makes it, and it flags the case where two agents are heading for the same symbol. Nothing reaches your main branch without clearing the gate.
-
How do I know what actually shipped?
Verdicts are derived from git: the commit on the branch, the diff, the tests. The agent’s own summary of its work is not on the list, because a self-report is not evidence. What cleared goes into a completion ledger that outlives the session, the harness and the machine, so next week you can still find out why something was built.
-
Can I really run it from my phone?
Yes. Post work, read the pipeline, approve a run and talk to Apollo from the phone app. The installer sets up a tunnel and a wake receiver so your phone can reach the desktop without you leaving anything open.
-
What does it cost?
Orbit is free while it’s in alpha, and anyone can sign up: your invite code is emailed right away. Pricing lands at beta. After alpha, new users get their first 14 days free.
-
Where does my code go?
Orbit runs on your own machine and your repositories stay there. Secrets live in a vault, so a run gets the keys it needs without you pasting them into a prompt. Whatever a prompt sends to a model goes to that provider on the terms of the key it used, and nowhere else.
-
How did Orbit start?
As a side tool budgeted at two to three hours. The first prompt, written at 1:19 in the morning on 26 March 2026, was a spec for something called Relay: a small local Node process that watched two files, the orchestration notes and the completion log, and rewrote a one-page status file, so a chat window could see what Claude Code was doing without a person copying summaries back and forth between them. It was read-only, made no API calls, and was meant to fit in one or two files. The completion log it watched is still at the centre of Orbit as the ledger. Everything else grew around the problem Relay was written for: the person should not have to be the glue between their agents.
The first prompt, 26 Mar 2026, 1:19 AM (excerpt) Relay — Spec v0.1 What it is: A lightweight local tool that gives Claude chat read-only visibility into an active Claude Code session, eliminating the manual relay step for solo founders using both tools together. Core problem: Claude chat (orchestrator) and Claude Code (executor) run in separate contexts with no shared state. The human copies prompts from chat → Code and summaries from Code → chat. Relay automates the visibility half. V1 scope: * File watcher on ORCHESTRATOR.md + COMPLETION_LOG.md * RELAY_STATUS.md generation * CLI: relay start, relay status, relay watch * No API calls, no external services — pure local file I/O Stack: Node.js, chokidar (file watching), commander (CLI). Single file or two files max. Runs as a background process. Fast build — maybe 2-3 hours.
Still deciding? Get your invite →