Hear the intent
Speak or type naturally. Operator preserves the original request and turns it into a durable task.
Operator by Xvoer
Operator is an AI assistant built to move from conversation to action. It can listen for intent, understand when people ask it to do something, and use Robot Mode to carry out authorized work.
Available today: Operator desktop, voice dictation, human steering, durable Robot Mode execution, approvals, evidence-backed memory. In development: Conference Mode and isolated virtual-computer execution.
ROBOT MODE
Most AI assistants stop at the answer.
How Operator works
Operator keeps the human request separate from execution, builds context from trusted evidence, asks when judgment is required, and carries authorized work through Robot Mode.
Speak or type naturally. Operator preserves the original request and turns it into a durable task.
Operator connects the request to local evidence, operating memory, project state, and prior decisions.
Robot Mode executes iteratively, pauses for important decisions, and applies approval gates to risky actions.
Results, decisions, command history, provenance, and evidence remain tied to the run instead of disappearing into a chat.
Conference Mode · In development
Conference Mode is the next step: invite Operator into a meeting, let it follow the conversation, and address it when you want work done. The goal is not another transcription bot. It is a participant that understands the request and can move the work forward.
Alex: We need the launch checklist updated.
Operator: I can handle that. I’ll inspect the current release state first.
Computer operation
Operator’s current Robot Mode can carry out iterative local work using the canonical desktop execution runtime, with durable recovery, human questions, steering, and safety gates.
The planned virtual-computer mode gives Operator its own environment so long-running work does not need to occupy your screen or interrupt what you are doing.
Longer term, dedicated Operator hardware can become a persistent, purpose-built computer for autonomous work — something you can add to a team like infrastructure.
Human control
The original human request is preserved independently of execution messages, so the system has a durable source of truth.
Higher-risk actions can require fresh approval tied to the exact action, run, and current steering state.
When a genuine product or business decision is needed, Operator can stop and ask instead of inventing the answer.
Durable sessions, command identity, recovery, and duplicate protection are designed to avoid uncertain mutations being repeated.
Operator is designed around local evidence and can use local models for selected transformations. Remote model use is explicit rather than silent.
Runs preserve decisions, evidence, execution history, and outcomes so completed work can be inspected instead of merely trusted.
What this changes
“Operator, reproduce that bug, inspect the relevant code, and prepare the fix.”
“Operator, update the action tracker from this discussion and flag anything without an owner.”
“Operator, verify what changed since last week and prepare the evidence for our review.”
“Operator, capture this request, check whether it already exists, and draft the follow-up.”
Recommended launch pricing
This is the recommended commercial model for early access. Final billing is not yet enabled.
INDIVIDUAL
For one person using Operator as a daily execution companion.
PRO
For people who want Operator continuously involved in serious work.
TEAM
For small teams building shared operating memory and delegated workflows.
Future hosted virtual-computer usage is expected to be metered separately because compute cost scales with workload. Dedicated Operator hardware will be priced separately when available.
Operator hardware · Longer-term vision
Software is the starting point. The longer-term product vision is dedicated Operator hardware: a quiet, always-available computer designed to run Operator continuously, join meetings, maintain context, and execute work without occupying your primary machine.
Hardware is a product direction, not a currently shipping product. Specifications, timing, and pricing have not been announced.
FAQ
No. Conference Mode is in development. The current product foundation includes Operator desktop, voice dictation, Robot Mode execution, durable questions and steering, evidence-backed memory, and approval mechanisms.
Operator’s current Robot Mode can execute iterative local work through its desktop automation runtime. The broader vision of generalized GUI operation and fully isolated virtual computers is still evolving and is not presented as generally available.
The system is designed around explicit intent, human steering, and action-specific safety policies. Higher-risk classes can require fresh human approval. The goal is useful autonomy with inspectable control.
The current macOS voice implementation uses explicit on-device dictation and does not retain audio files. Conference Mode architecture and privacy details will be finalized before release.
Operator can use local Ollama models for selected work and supports explicit frontier-model routes. Its architecture is designed so local evidence is not silently sent to a different remote provider.
Not yet. Dedicated hardware is part of the longer-term vision and is shown here to make the direction of the product clear, not as a shipping SKU.
Early access
Operator is under active development. Early access is for people who want to help shape how AI moves from conversation to real work.
Request early access No payment required. Commercial plans shown above are recommended launch pricing and may change before billing begins.