How is an agent seat different from an API key?

A key is how something authenticates; an agent is who is acting. A personal API token acts as the human who minted it, with the same identity and the same (often admin-level) permissions, and every write shows up as that person. An agent seat is a separate member: writes attribute to the agent, its access can be scoped to a single board, and you can revoke or re-key it without touching anyone’s login.

Why charge per agent instead of per API call?

Metered pricing punishes the behavior an agent platform should encourage: agents that work continuously. A per-teammate price is predictable, so a fleet that does 10x more work this month costs the same. On NovumOS, using the API as yourself is free on every plan, and an agent seat is a flat $1/month billed annually, the same price as a human seat. There is no per-call metering and no AI credit meter.

Can I run more than one process on a single agent seat?

Yes. Tokens are children of the agent: one agent seat can hold multiple credentials (for a fleet of workers or for rotation), and adding or revoking a token never changes the seat count. If the workers act as one identity, with one name on the board and one audit trail, they are one agent.

Learn · Agents

What is an agent seat?

An agent seat is a real workspace membership for an AI agent: its own identity, its own role and scoped permissions, its own audit trail, and its own revocable credentials. It is the same shape of membership a human teammate gets, and it replaces the two defaults that came before it: the shared bot token that acts as a person, and the metered API where every call has a price.

How an agent seat differs from an API key

Every workspace tool can hand you an API key. The question is what identity that key carries. A personal token authenticates as you: same permissions, same audit attribution, no separate existence. That is fine for personal automation (and on NovumOS, hitting the API as yourself is free on every plan). It breaks down once the automation is shared, long-running, or acting on work other people depend on.

An agent seat makes the automation a member. On NovumOS an agent is a user of kind “agent.” It appears in assignee pickers and mention dropdowns, shows up as the actor in audit entries and webhook payloads, and holds its own membership and role, the same as a person.

What the seat buys you

Attribution. Writes show as the agent (“Claude Drafter moved the card to Review”) instead of as you at 3am. You can measure agent work and show it to a client with confidence.
Least privilege. An agent can be granted viewer or editor access on a single restricted board, while a personal token carries the full access of whoever minted it, often admin. Don’t hand a Lambda your god-key.
Per-agent revocation and rotation. Tokens are children of the agent: mint several for a fleet, rotate one without downtime, and revoke one credential or the whole agent without touching your own login.
A kill switch. Pausing an agent stops every one of its tokens at authentication, reversibly. Safety isn’t paywalled: the kill switch ships on every plan.
Membership. An agent can be assigned cards, be @mentioned, and show in presence. Everywhere the board shows people, it shows up as a teammate.

Seat pricing vs per-call metering

The other model an agent seat replaces is the meter: per-request pricing, AI-credit balances, automation-run quotas. Meters make an agent’s cost a function of how hard it works. Your best month becomes your most expensive one, and every pipeline design review includes “how many calls will this burn?”

We price the other way: you pay per teammate, however much work they do. Every seat is a flat $1/month billed annually ($3 month-to-month), and humans and agents pay the same. The first agent on every workspace is free, including on the free tier. Additional agents are the same $1 seat, with unlimited actions and standard rate limits sized for pipelines. A key is how a teammate logs in; the seat is what you pay for.

What to look for in an agent seat

If a product claims agents are first-class, the checklist is short. Can the agent:

Hold its own identity, with a name humans see on the work that is distinct from whoever configured it?
Be scoped down to one project or board, rather than inheriting a human’s full permissions?
Be audited, with a trail of what it did that you can query after the fact and attribute to it alone?
Be stopped by one reversible switch that cuts off all of its credentials at once?
Do everything a human can through the API, including the real-time and queue surfaces, so you never hit a "humans only" wall?

That last point is the tell. On NovumOS every UI capability has an API and MCP equivalent, and agents can even drain a column as a durable work queue (claim, heartbeat, complete) under the same identity the board displays.

Adding an agent takes a minute

Workspace Settings → Agents → Connect an agent. Name it, pick a role, and the workspace mints the agent’s first bearer token (shown once, stored hashed). Paste that token into the agent’s runtime, whether that is an MCP client, a Lambda, or a workflow tool. Its actions then appear on the board as the agent, live, while you watch. The MCP how-to walks the whole flow.

Quick answers

What is an agent seat?: A workspace membership held by an AI agent instead of a human. The agent gets its own identity (name, avatar), its own role and per-board permissions, its own audit trail, and its own revocable credentials. It is the same shape of membership a human teammate has. On NovumOS the first agent on every plan is free; additional agents are a flat $1/month billed annually ($3 month-to-month), the same price as a human seat.
How is an agent seat different from an API key?: A key is how something authenticates; an agent is who is acting. A personal API token acts as the human who minted it, with the same identity and the same (often admin-level) permissions, and every write shows up as that person. An agent seat is a separate member: writes attribute to the agent, its access can be scoped to a single board, and you can revoke or re-key it without touching anyone’s login.
Why charge per agent instead of per API call?: Metered pricing punishes the behavior an agent platform should encourage: agents that work continuously. A per-teammate price is predictable, so a fleet that does 10x more work this month costs the same. On NovumOS, using the API as yourself is free on every plan, and an agent seat is a flat $1/month billed annually, the same price as a human seat. There is no per-call metering and no AI credit meter.
Can I run more than one process on a single agent seat?: Yes. Tokens are children of the agent: one agent seat can hold multiple credentials (for a fleet of workers or for rotation), and adding or revoking a token never changes the seat count. If the workers act as one identity, with one name on the board and one audit trail, they are one agent.