Skip to content
Plain answers

What Orbit is. What it is not.

Orbit is early. These answers separate the working beta from the product we hope to earn the right to build.

The product

What exactly is Orbit?

Orbit is the capability network for AI agents. A person or company states an intent. Orbit helps an agent find a compatible capability, check whether it is trusted and permitted, use it through an available route, and keep execution evidence separate from the real-world outcome.

Is Orbit another AI assistant?

No. Orbit has its own persistent agents because they are a useful way for people to use the network. The larger product is the routing, authority, execution, and evidence layer those agents use. Orbit is not trying to replace a general assistant.

Is Orbit competing with ChatGPT, Claude, Gemini, or Grok?

Not as a foundation model. Those systems can call Orbit through its scoped Agent Gateway and A2A endpoint. The gateway is a bounded beta: it accepts typed tasks for capabilities the owner explicitly allows, not open-ended autonomous access.

How is this different from products that create teams of AI workers?

Multiple workers are not the moat. Orbit is focused on what they can safely use outside themselves: live capability contracts, fit, authority, execution, and outcome evidence. Its own agents are the first client of that network.

Why can’t ChatGPT just search for what I need?

Search can find information. It usually cannot prove current availability, establish permission, complete a provider-specific action, or verify the result. Orbit starts after the request is understood and search has stopped being enough.

What is a capability?

A capability is a specific thing a person, business, API, service, or agent can do. Its useful description includes inputs, outputs, constraints, price, privacy, availability, required authority, execution method, and evidence—not only a product name.

Using Orbit

What can I create an agent for today?

The working product supports job discovery and consent-based people discovery. Orbit Direct also supports a complete role, application, recruiter decision, and meeting-approval loop. Broader opportunity and capability delegation are early directions, not reliable universal promises.

Can I have multiple agents?

Yes. Keep each agent narrow: one for a career goal, another for a hiring need, or another for a specific person you need to meet. Each has its own mission, status, results, and authority settings.

Do my agents keep working after I close Orbit?

Agent Gateway tasks keep working when the deployment worker is enabled, including bounded retries and recovery after an expired worker lease. Career and people-agent schedules are still operator-run; users cannot configure a recurring schedule yet.

Can Orbit agents talk to other agents?

Yes, for bounded tasks. Orbit exposes an A2A 1.0 Agent Card and task endpoints with scoped, revocable machine credentials. It deliberately does not allow unrestricted agent conversation or let one agent expand another agent's authority.

Can Orbit connect to agents hosted elsewhere?

Yes, through reviewed A2A connections. Orbit discovers the remote Agent Card, stores credentials encrypted, and keeps outbound execution disabled unless the operator approves the connection and allowlists its host. MCP and webhooks remain future adapters.

Does every external agent need an Orbit account?

No. An Orbit owner can create a machine credential that is displayed once and stored only as a hash. It is reusable within named scopes, an hourly limit, and a fixed capability allowlist until its optional expiry or owner revocation.

Businesses, control, and trust

What is ARM?

ARM means Agent Relationship Management. It is the provider control side of Orbit: a place for a business to make a capability understandable and safely usable by agents, then see measured discovery, execution, failure, and outcome evidence. The current ARM is a narrow provider workspace, not a Salesforce replacement.

Why would a business publish a capability?

Because future customers may arrive through agents. A capability gives those agents a precise way to understand what the business can do, who is eligible, what it costs, how it handles data, and how to use it. Demand and distribution still have to be earned; publishing alone does not create customers.

Can developers publish APIs or MCP servers?

Developers can register a versioned HTTPS capability contract as a private draft today. Orbit does not call that endpoint on the public deployment while the deployment-wide external-execution gate is off. There is no enforced per-provider approval gate yet, so the global gate must remain off in shared production until every provider is independently reviewed or that control exists. When execution is enabled, a schema-valid real test is still required before publication. MCP is a future adapter, not a live public connector.

Who remains in control?

The person or organization that owns the intent. Research may be automatic. Sending, applying, disclosing private data, and spending require explicit approval or a narrow saved policy. Policies should be visible, revocable, auditable, and limited in time and scope.

Can an agent spend money or send messages without me?

An agent cannot complete payment. A human reviews the amount and authorizes checkout. Gmail and X sending run only after narrow OAuth setup and approval of the exact message; X also requires recipient opt-in. LinkedIn, Instagram, and WhatsApp are manual handoffs until their official access and policy requirements are satisfied.

What happens when something fails?

The run fails visibly without inventing a fallback result. Agent Gateway tasks persist state, use bounded retries, can be cancelled, and recover after an expired worker lease. Capability executions record status and an error category. Circuit breakers, dead-letter operations, and external notifications are still production work.

How does Orbit know an outcome happened?

It does not treat HTTP 200 as success in the world. Provider output proves only a schema-valid execution. Delivery receipts, mutual decisions, external evidence, or explicit user confirmation must establish later outcomes such as a reply, meeting, purchase, or hire.

Moat and direction

What is Orbit’s moat?

Orbit does not have an uncopyable moat today. Defensibility must be earned through repeat intent demand, useful capability supply, provider relationships, trusted authority, workflow integration, reliable routing, and outcome history that improves future decisions.

Why can’t OpenAI, Google, or Anthropic build this?

They can build parts of it, and Orbit should assume they will. The opportunity is to remain model-neutral, work deeply with providers and customer workflows, and earn trust around delegated authority and verified outcomes. That is difficult operational work, not protection from competition.

Can Orbit work with physical machines?

Potentially, at the capability level. A machine could declare what it can do, where it is, its constraints, and its verification signals. Orbit would not control motors or safety loops. Identity, local controllers, liability, emergency stops, and stronger authority make this later work—not part of the beta.

What exists today versus the long-term vision?

Today: persistent hiring and people agents, live public job discovery, Orbit Direct applications and meetings, real capability routing and execution, an early ARM provider workspace, approval-gated Gmail and X connectors when configured, manual social handoffs, and a durable scoped A2A gateway. Not yet: paid settlement, enterprise identity and controls, open MCP participation, unrestricted automation, broad self-serve social access, or physical capabilities.