Orbit
General operator · routes the work
Features
Five specialists. One instruction. Zero tabs.
Orbit isn't a chatbot bolted onto your desktop. It's an agentic operating system that reads, clicks, types, and ships — every action visible, every action gated. Here's exactly what it does.
Free forever · Bring your own API keys · Local-first.
Multi-agent automation
Five specialists that actually work together.
One instruction fans out to a team. Orbit plans the job, hands each step to the right specialist, and stitches the results back together — no tab-hopping, no copy-paste, no babysitting. You describe the outcome; the swarm delivers it.
- Orbit orchestrates: decomposes goals and routes work to the right agent.
- Researcher gathers, reads, and synthesizes sources into a brief.
- Coder reads repos, writes patches, runs tests, and proves the fix.
- Operator drives apps, files, and the OS to complete real tasks.
- Writer drafts, edits, and formats the final deliverable.
- Agents pass context to one another — a research brief becomes a written report becomes a saved file, in one run.
Researcher
Finds, reads, synthesizes
Coder
Writes, runs, and fixes code
Operator
Drives apps, files, and the OS
Writer
Drafts and polishes the output
live · operator
RUN rename & file 214 invoices
✓ read Downloads/ — 214 matches
✓ browser: fetched vendor names
✓ shell: mkdir 2026-{01..12}
→ moved 214 files · 0 errors
Deep computer & browser automation
It operates the machine — not just the chat box.
Orbit reaches every layer of your desktop: the filesystem, native apps, a real headless-capable browser, and the shell. These are genuine tool calls on your computer, executed in the open and reversible by design.
- Files: search, read, rename, move, organize, and batch-transform at scale.
- Apps: launch and drive native applications to finish real workflows.
- Browser: a dedicated Playwright profile navigates, clicks, fills forms, and extracts data.
- Shell: runs commands, scripts, and build steps — each gated by your approval.
- Code: reads repositories, edits files, runs the test suite, and re-runs to prove the fix.
Trust & permissions
Autonomy you can stop with one key.
Power without oversight is just risk. Orbit is built the other way around: every capability is tiered, every action is logged, and a single keystroke freezes the entire fleet. Nothing runs in the dark, and nothing runs without your say-so.
- Permission tiers: read, act, and destructive — each scoped per agent.
- Consent prompts: every new action asks the first time; every command asks every time.
- Full audit log: a human-readable record of every tool call — what, when, and the result.
- One-key kill switch: freeze all agents instantly, or hold the Orb to do the same.
- Revoke anytime: grants are yours to take back, per agent and per capability.
audit log
✓ read file · Downloads/q3.csv · granted
✓ act · move 14 files · granted
● pending · run npm test · awaiting consent
KILL Ctrl+Shift+Esc — all agents frozen
3Permission tiers
1-keyKill switch
100%Actions logged
0Bytes sent by default
On-deviceMemory store
Your projects, preferences, and history live in a local vector store on your disk — recalled instantly, never uploaded unless you turn on cloud sync.
Local-first memory & knowledge
It remembers your world — on your machine.
Orbit builds a private memory of how you work: the folders you care about, the way you name things, the projects in flight. That context makes every run sharper — and it stays on your device unless you decide otherwise.
- On-device vector memory — no data leaves your machine by default.
- Persistent context: preferences, project state, and past runs recalled automatically.
- Bring-your-own knowledge: point agents at local folders and docs.
- Optional cloud sync (Pro) — opt-in, encrypted, and yours to disable.
Voice control
Say it. Watch it happen.
Talk to your computer like you'd talk to a capable teammate. Voice input runs on-device, kicks off the same permission-gated agent flow, and lets you drive a full workflow without touching the keyboard.
- Hands-free commands trigger the full multi-agent pipeline.
- On-device speech recognition — audio stays local.
- Speak a goal, confirm the plan, and watch the Orb execute.
- Push-to-talk or wake-word — your choice.
you said
"Pull last week's sign-ups and draft the summary."
Researcher exporting sign-ups…
Writer drafting summary…
✓ summary.md saved
The reactive Orb & cinematic UI
A single presence that shows you the work.
The Orb is Orbit's face — a living pulse that breathes with activity, tints to the agent currently working, and turns into your kill switch when you hold it. The whole interface is built to make autonomy legible, not mysterious.
- Reactive Orb: pulses on activity, tints to the active agent's accent color.
- Live action feed: every step streams in as it happens.
- Hold-to-stop: press and hold the Orb to freeze the fleet.
- Premium dark UI — restrained neon, buttery motion, zero clutter.
Model routing
The right model for every step.
Not every task needs the biggest model. Orbit routes each step to the Claude model that fits — reserving deep reasoning for hard problems and fast, cheap inference for the rest. You get better results and a lower bill, automatically.
- Opus for complex planning, hard reasoning, and multi-step orchestration.
- Sonnet for everyday tasks — the balanced default for most work.
- Haiku for fast, high-volume steps and quick classification.
- Automatic routing per step — no manual model-picking required.
- Bring your own API keys; Pro adds priority models and higher throughput.
routing
Opus plan & decompose goal
Sonnet read repo · write patch
Haiku classify 900 files
→ cost-optimized · 1 instruction
Get started
It doesn't answer. It works.
Free forever. Local-first. Your data stays on your machine.