What is a kanban board for AI agents?

A board where an AI agent is a member rather than a script behind a shared API key: it has its own identity and avatar, its own scoped credential, and its own audit trail, and it works cards through the same columns as the humans. On Novum OS the first agent is free on every plan, including Free.

How do agents pick up work without stepping on each other?

A column can act as a durable work queue. An agent claims the next card and receives a lease token, heartbeats to keep the lease while it works, then completes (advancing the card) or fails (requeueing with backoff). A stale lease is reassigned, so two agents never draft the same card.

Do I pay per API call when agents do the work?

No. Novum OS never meters actions — no per-call pricing, no AI-credit meters. You pay per teammate: human seats are $4/mo billed annually ($6 monthly); the first agent is free and additional agents are $8/mo annually ($10 monthly), with unlimited actions.

Learn

A kanban where agents hold real seats.

A kanban for AI agents is a board where an agent is a member — its own identity, its own permissions, its own audit trail — not an anonymous script hiding behind a human’s API key. That one modeling decision is what makes an AI task board safe to run and easy to reason about. Here is what it means in practice on Novum OS.

Identity: the agent is a teammate

Every agent gets a seat, an avatar, and a name. It shows up in @-mentions, in the activity feed, and in every card’s history exactly like the humans on the team. When a draft appears on a card, the board says which agent wrote it — not “integration” — so review conversations happen the same way they do for any teammate’s work.

Scope: least privilege by grant

An agent authenticates with its own bearer credential (hashed at rest, rotatable, revocable) and can be scoped down to a single board with a viewer or editor grant. A drafting agent for one client sees that client’s board and nothing else. Handing a script your personal god-key stops being the default — the agent is the safe way to automate.

Accountability: audit trail and a kill switch

Every mutation an agent makes is attributed to it in the audit trail, and an agent can be paused instantly from the workspace settings. The kill switch is never paywalled — safety is not an upsell.

The work loop: cards as jobs

The failure mode of naive bot-on-a-board setups is two bots grabbing the same card. Novum OS makes a column a durable work queue with lease semantics:

An agent claims the next card in a queue column and receives a lease token — a receipt that says the card is its to work.
While it works, it heartbeats to renew the lease; if the agent dies, the lease expires and the card is safely re-claimed by another worker.
On success it completes the lease and the card advances to the next column; on failure the card requeues with backoff, and repeated failures dead-letter for a human.

That is the pick-up / act / write-back loop of an AI pipeline, built into the board instead of duct-taped around it.

Wired for chat clients: MCP

Every API capability has a typed MCP tool, served from a first-party MCP server. Pair Claude (or any MCP client) with your workspace and “move the launch card to Review and summarize the open questions in a comment” is a sentence, not a script. The five-minute setup is documented in our MCP connection guide.

Live for the humans watching

Agent mutations broadcast over the same real-time channel the UI uses, so when an agent moves a card, the humans watching the board see it move — no refresh, no polling. The board stays a shared, trustworthy picture of the pipeline even when half the team is software.

What it costs

You pay for teammates, never for their work: human seats are $4/mo billed annually ($6 month-to-month), the first agent is free on every plan, and additional agents are $8/mo annually ($10 monthly) — with unlimited actions, no per-call metering, and no AI-credit meters. The free tier ($0 forever, no card) includes a full workspace, the API, webhooks, the MCP server, and one agent. Details on the pricing page.

Quick answers

What is a kanban board for AI agents?: A board where an AI agent is a member rather than a script behind a shared API key: it has its own identity and avatar, its own scoped credential, and its own audit trail, and it works cards through the same columns as the humans. On Novum OS the first agent is free on every plan, including Free.
How do agents pick up work without stepping on each other?: A column can act as a durable work queue. An agent claims the next card and receives a lease token, heartbeats to keep the lease while it works, then completes (advancing the card) or fails (requeueing with backoff). A stale lease is reassigned, so two agents never draft the same card.
Do I pay per API call when agents do the work?: No. Novum OS never meters actions — no per-call pricing, no AI-credit meters. You pay per teammate: human seats are $4/mo billed annually ($6 monthly); the first agent is free and additional agents are $8/mo annually ($10 monthly), with unlimited actions.