Orbit is a free desktop app during alpha for developers who use AI coding agents like Claude Code and Codex. Your invite code lands in your inbox in about a minute.
Everyone else is building better workers — for their own model. Who is building the workplace? Orbit gives Claude Code, Codex, and Gemini desks of their own, a job board to pull from, an approval gate on the way out, and a permanent record of what actually shipped — verified against git, not taken on the agent's word. Work routes across the subscriptions you already pay for, or your own hardware for free, and you see the bill either way.
It all runs on your machines — and you run the whole office from your phone.
20+ features shipped
●Invite code in your inbox in a minute
Windows · macOS · Ubuntu
your phone · tap to approve
▎ the shop floor
A good office doesn't do the employees' work.
The harnesses own the intelligence, and they're getting better every month — parallel agents, subagent swarms, longer context. Let them. What none of them owns is the thing that sits above any one session and any one vendor: the durable queue, the gate, the receipts. Orbit isn't trying to be a smarter agent...it's the building they clock into.
arrow_downward scroll to walk the floor · click anything
01 / 08Ask ApolloSay what's wrong in plain words. Apollo drafts the prompt, and when you say queue it, the work order prints and goes up on the board. Planning is a conversation, not a form.
02 / 08The job boardOne durable queue across every project and every machine. Post work from your phone at 6am; it’s still there, in order, when a worker is free.
03 / 08The employeesClaude Code, Codex, Gemini — whichever company they come from, they work here. They get smarter on someone else’s budget, and you keep the upgrade.
04 / 08The desksEvery run gets its own git worktree. Nobody works on top of anyone else, and nobody has to be told to wait their turn.
05 / 08The inspectionVerdicts are derived from git — commits, diffs, what’s actually on the branch — not from the agent’s summary of its own work. Self-reports don’t clear the door.
06 / 08The work orderNothing starts on a hunch and nothing lands unsigned. Approval gates sit at the front of the run and at the door on the way out.
07 / 08The recordA completion ledger that outlives the session, the harness, and the machine. When something breaks next week, the reason it was built is still here.
08 / 08The daily summaryAt the end of the shift you get one page, written for you: what shipped, what was sent back, what is still at a desk. You never read a log.
arrow_downwardScroll down to move through the office →
Loading the office
what you're looking at
01
Ask ApolloSay what's wrong in plain words. Apollo drafts the prompt, and when you say queue it, the work order prints and goes up on the board. Planning is a conversation, not a form.
02
The job boardOne durable queue across every project and every machine. Post work from your phone at 6am; it’s still there, in order, when a worker is free.
03
The employeesClaude Code, Codex, Gemini — whichever company they come from, they work here. They get smarter on someone else’s budget, and you keep the upgrade.
04
The desksEvery run gets its own git worktree. Nobody works on top of anyone else, and nobody has to be told to wait their turn.
05
The inspectionVerdicts are derived from git — commits, diffs, what’s actually on the branch — not from the agent’s summary of its own work. Self-reports don’t clear the door.
06
The work orderNothing starts on a hunch and nothing lands unsigned. Approval gates sit at the front of the run and at the door on the way out.
07
The recordA completion ledger that outlives the session, the harness, and the machine. When something breaks next week, the reason it was built is still here.
08
The daily summaryAt the end of the shift you get one page, written for you: what shipped, what was sent back, what is still at a desk. You never read a log.
6:47 PM. You're not at your desk. A user reports a crash on iOS. You open Orbit on your phone, tell Apollo what's broken. He queues the fix to Claude Code on your PC. It runs, passes QA, builds via Fastlane on your Mac. You approve the TestFlight submission from the couch. Done before dinner.
6,227commits that cite the prompt that made themsince March 26, 2026
4,630runs the ledger marks shippedverified against git
834runs it marks partial or failedthose stay in the ledger too
Orbit is not a smarter agent. It is the building the agents clock into, and these are the parts of your week it takes off your hands.
01
Queue it and walk away
Stack work across every project from wherever you are. Run it in order or in parallel. Queue twenty-five prompts before bed and read the ledger in the morning: Orbit dispatches them, holds the ones that need you, and keeps going.
from the ledger, this week
#7066Office hero: the room on orbitdev.app✓ shipped
#7057Identify provider and quota window in usage alerts✓ shipped
#7037Fable usage meter in the usage panel✓ shipped
Claude Code, Codex, Gemini, Grok, DeepSeek, Kimi, GLM, Llama, anything OpenRouter fronts — and anything else OpenCode can run, under whichever harness you prefer. Same queue, same context, same secrets. Swap the model and keep the workflow.
who clocks in
Claude Code
Codex
Gemini
Grok
DeepSeek
Kimi
GLM
Llama
OpenRouter
One queue. One vault. One ledger. The model is a setting.
Two windows, side by side. Chat on the left, Code on the right. Neither one can see what the other is doing, so you copy the context out of one and paste it into the other. Again. You're not building anything...you're just a translator between two coworkers who refuse to sit at the same table.
03
Nothing ships on its own say-so
A run's verdict comes from git: the commits, the diff, what is actually on the branch. The agent's summary of its own work is not evidence. If it didn't land, the ledger says so, in writing, forever.
a real receipt
prompt#7057 · Identify provider and quota window in usage alerts
commit262d72465
touchedsrc/usage-alerts.js · src/dashboard.html
verdict✓ shipped · diff matches report · no self-report used
834 runs are marked partial or failed. They stay in the ledger with the same detail.
04
Run it from your phone
Post work at six in the morning. Approve it from the couch. Dispatch a build to iOS, Android, Windows, or Mac from the school pickup line. The office does not need you in the room, it needs you to sign.
what your phone shows
6:47●●●
ORBIT · now
#7066 ready for review
Office hero · claude · +654 −261 · build passed
✓ diff verified against git
ApproveHold
#7061shipped
#7060shipped
#7058in review
#7052queued
Five projects, and your attention is the thing that runs out first. Each one has its own queue, its own half-finished thread, its own "wait, where did I leave this." Orbit puts all five on one screen — every queue, every agent, every completion log, in a single feed. Queue ten prompts across all of them and let the approval gates and QA checks hold the line while you're somewhere else. You stop holding it together in your head...you just watch it ship.
05
The project remembers
Your CLAUDE.md and AGENTS.md still work. Orbit adds an orbit.md that every agent gets, whatever the model, plus the Memtrace code graph and a wiki the agents write for each other. One memory, handed to whoever runs.
one memory, every model
CLAUDE.mdAGENTS.mdGEMINI.md.cursorrules
→
one project memoryconventions · gotchas · decisions · the code graph
Injected into every prompt before the agent reads a line. Nobody has to remember to @-mention it.
06
Security stops being a someday job
Run a security audit on any project without leaving Orbit. Local checks look for leaked secrets and dependencies with known vulnerabilities, at no model cost, and a deep audit adds static analysis and a model review. Findings come back in plain language, and a fix is a proposal you approve that gets verified like any other change.
what an audit hands back
Security · orbitquick check · local
highSaved secret in sourceserver credential written as a literalfix proposed
mediumKnown vulnerability (CVE)dependency pinned below the patched releasepin ready
Nothing is applied until you approve it. The fix runs through the same gate and the same git check as every other change.
You write the gotcha down once; the convention, the reason this file is weird, the thing that broke last time. But one model wants it in CLAUDE.md...the next wants AGENTS.md...and then a third keeps its own. So you maintain the same memory in three places and hope the right copy loaded before an agent does something dumb. Orbit holds it once and hands it to whatever model runs. The project remembers, so you don't have to.
▎ also in the box
And the rest, quietly doing work.
Built because we needed them. Each one has a page.
Alpha runs today, and anyone can get in. 1.0 ships when we'd hand it to a friend without caveats.
NOW · α.4
Alpha · open signup
Everything above runs today on Windows, macOS and Ubuntu (Linux). Daily ships. Sign up and your invite code arrives by email right away. free while in alpha.
YOU ARE HERE
NEXT · β.0
Beta · pricing arrives
Paid tiers arrive (see pricing), and new users get their first 14 days free. Everything in alpha today carries over: the pipeline, Memtrace, Sivants Studio, sandbox sessions, and the phone app.
NEXT
LATER · 1.0
Launch · general availability
Cloud sync, on-call agent profiles. Org plan with SSO & audit log. BSL → Apache 2.0 countdown begins.
AFTER BETA
There is no line.
Alpha is free and ships daily. Drop your email and your invite code is in your inbox in about a minute: download, sign in, paste the code, and you're working.
No queue · code sent right away · free while in alpha.
▸ what happens nextabout 2 min
1 check your inboxYour invite code arrives right away, from chris@sivants.com
2 download Orbitorbitdev.app/install · Windows 10+, Apple Silicon Mac, or Ubuntu via the Linux helper
3 sign in, paste the codeGitHub or Google. Apollo walks you through your first project.
