Questions, answered without rounding anything up

UtilitySpaces is an agentic workspace: a chat-first inbox shared by the people on your team and the AI agents working alongside them. You assign work by writing a message, agents act on it using scoped access to the business applications you have connected, a connector write waits for a person to approve it unless an admin has set that connection to run without asking, and every call is written to an append-only audit log.

Teams that already run their operations across several applications and have recurring work that nobody enjoys: order triage, inbound enquiry qualification, weekly reporting. It assumes you have real systems worth connecting. If your business runs entirely out of one inbox and a notebook, you do not need this yet.

Yes, and there is no other way to try it, because there is no salesperson. You open a demo workspace from the sign-in page with no card and no call. Nobody will phone you afterwards.

A throwaway demo lives for exactly one hour from the moment it is created, then it is expired and its contents are removed. It is a throwaway space on purpose: it cannot invite other people, and it is limited to 5 agents, 10 agent tasks and 1 standing schedule. Create an account from it before the hour is up and the clock stops entirely — the workspace keeps everything you built and stays free at those same limits until you choose a paid plan.

Connecting one application and giving one agent one job is a job for an afternoon, not a quarter. Getting an agent you trust enough to stop watching takes longer, and it should: the honest sequence is one job, one agent, one week of reading every run before you widen anything. There is a longer argument about which job to pick first in what to automate first with AI agents.

Yes. The throwaway demo you open from the sign-in page lives one hour, but creating an account from it stops that clock: the workspace then keeps working at demo limits — 5 agents, 10 agent tasks, 3 seats — until you choose a plan. What waits on an invoice is the paid ceilings, not the workspace.

Starter is $29 per month, Team is $79 and Business is $249. Enterprise is scoped and priced around your own workload, inside a published band of $499 to $4,999 per month — its ceilings are set by the quotation rather than by a tier, invoicing terms are agreed before the first invoice, and nothing is committed until the quotation you sign says it. Model usage is not part of any subscription: on a paid plan you bring your own provider key and pay your provider directly, and only the free demo runs on models we supply. The pricing page lists what each tier includes, and the Enterprise pricing guideline (PDF) sets out what moves a quotation inside that band.

Paid plans on UtilitySpaces are monthly subscriptions paid by card. Checkout runs through our payment provider, acting as the merchant of record, so your card number never reaches us; the plan activates within a minute and renews automatically on the same day each month until you cancel. Visa, Mastercard and local methods are accepted worldwide, and any tax due in your country is added at checkout. If a renewal is ever declined nothing is switched off — we tell you and the card is retried. Cancel any time from workspace settings; you keep access until the end of the month you paid for. Billing questions go to support@utilityspaces.com.

Two things. Standing schedules: a recurring schedule keeps producing autonomous work without anyone asking for it, so it is the resource the tiers are written against — 5 on Starter, 15 on Team, 40 on Business. And hosted execution: the agent tasks a month our shared brain will run for you — 300 on Starter, 1500 on Team, 5000 on Business. Work you run on a brain of your own is not counted and is not capped at all. Chatting to an agent and giving it one-off tasks are deliberately not metered on any paid tier; seats, agents and attachment storage have ceilings too, but those two are what a tier is sold by.

The workspace stops running paid work. Your data is not deleted the moment an invoice is late; it is retained so that you can export it or settle up. If you want it gone rather than dormant, the data deletion page tells you exactly how to ask and what it removes. The refund policy covers the money side.

A workspace owner can export the whole workspace from Settings at any time, as a download that leaves out stored credentials, and you can ask for it to be deleted. Deleting a workspace moves it to a recycle bin, and when that window ends its contents are purged. Two things are kept after the purge: invoices, and the workspace's audit log, which is the tamper-evident record of what was done in the workspace, including anything done by us. The log stays attached to a marker that the workspace existed, and no scheduled job deletes it. The details are on the data deletion page and in the privacy policy.

Yes. Sign-in supports TOTP from any authenticator app, with ten single-use recovery codes, and an owner or admin can require it for the whole workspace. The security page lists what is and is not in place without rounding any of it up.

Yes. OpenID Connect against your identity provider — Entra ID, Okta, Google Workspace and anything else that publishes a discovery document — configured per workspace against a domain you verify by DNS. Accounts are provisioned on first sign-in. An owner can require single sign-on for that domain while keeping a break-glass owner who can still sign in with a password. SAML and SCIM are not supported.

No. There is no third-party security certification. If your procurement process requires a SOC 2 or ISO 27001 report, UtilitySpaces will fail it, and the honest answer is that the managed service cannot meet that requirement today. If that is what is blocking you, say so rather than working around it — knowing which requirement blocked a real team is how the order of this work gets decided.

There is a hash-chained, append-only audit log. Each entry carries a sequence number, the hash of the previous entry and its own hash, so editing or removing any entry breaks the chain at a known point and a verification pass reports exactly where. What that proves is that the log has not been altered since it was written — it does not prove the log is complete, and no hash chain does. There is more on how the record is used in the piece on designing the approval gate.

An agent reaches exactly the applications you have connected, through that connector’s own tools and, for OAuth connectors, that connector’s own scopes. Each connection is a separate permission boundary; nothing is shared between them by default. Reads are not gated. Connector writes wait for a person to approve them unless an admin has set that connection to run without asking, because a wrong read is recoverable and a wrong write is durable. Every outbound request is checked against a guard that refuses private, loopback and internal network addresses on every redirect hop — the reasoning is set out in the four ways to give an agent app access.

Only if an admin has allowed it. A message sent or money spent through a connector waits in Approvals for an owner or admin, unless an admin has set that connection to run without asking; Chatwoot and WAHA connections that ran unattended before approvals reached connectors were kept that way, and every change to that setting is logged. A delete always waits for a person. Message drafts and document updates inside the workspace raise an approval request that names the agent, shows the actual content about to go out, and waits for a person; a rejection sends the work back to the agent rather than throwing it away.

On a paid plan you bring your own. A workspace stores 1 provider credential — the same on every tier, because the store is keyed by workspace — and it can point at any OpenAI-compatible or Anthropic-shaped endpoint reachable over public HTTPS. You pay your provider directly; model usage is not part of the subscription. The free demo is the exception: it has no key of its own and runs on the shared preview brain, on models we supply. Either way the product records what each run cost, by model and by workspace.

Three things, and none is magic. Standing schedules are capped per plan, so the amount of unattended recurring work a workspace can create has a ceiling. Hosted execution is capped per month as well, so runaway work on our shared brain stops rather than compounding — though work on a brain of your own is uncapped, and there the ceiling is your own hardware. And a write capability resolves to exactly one destination per workspace: if two connected applications could both service a write and nobody has chosen between them, the action is refused rather than guessed at. Volume anomalies still need a human reading the log, which is why the log is in the same place as the work.