Plans and credits
There are two separate things to understand: the plan, which decides which features, workspaces, seats and resource limits you get, and credits, which measure AI usage. Limits and billing apply to the account, not to an individual workspace.
For current prices, seat counts and credit allowances, see the pricing page — it is the authoritative list. This page explains how the system behaves.
Plans
Three plans are available self-serve, plus Enterprise by arrangement:
- Free — build, upload data, create agents, chat, evaluate and publish to the client Portal. Specific connectors and advanced capabilities still follow the feature matrix.
- Pro — more workspaces, storage, datasets, scheduled jobs, client seats and monthly credits.
- Team — higher resource limits and collaborator seats; some resources may be uncapped.
Plan limits are aggregated across the account rather than granted separately to each workspace. Workspace count, storage, datasets, scheduled jobs and online Portals are counted across the account; seats are counted by unique person. Dataset version retention depth also depends on the plan.
Publishing is not a paid feature. Every plan can publish a Portal. The account's structural capacity determines how many workspaces and online Portals it can have, with one Portal per workspace. Adding agents to an existing Portal does not consume another Portal allowance.
Other features are gated by plan, not by workaround. When one is out of reach, the app returns a clear upgrade prompt rather than failing silently. Downgrading does not delete existing workspaces, datasets, scheduled jobs or Portals; the account may temporarily exceed its new limits. You can still read, export and delete existing content, while creating resources or starting new refreshes may be blocked. Taking a Portal offline is always free.
Credits
Credits measure AI work. They are not one credit per question: model calls and media generation are measured by what they actually consume.
For language and speech models, usage is derived from tokens: the input and output tokens of each call are priced at that model's rate and converted to credits. Media generation also uses applicable units such as rendered video or spoken audio duration, document pages or slides, and image count. Model rates are maintained by platform administrators and may change with provider pricing. Consequences worth knowing:
- A long document in context costs more than a short question, even for the same answer.
- A cheaper model costs fewer credits for the same work. Agents do not all use the same model.
- Failed calls are not charged. Reserved credits are released or settle at zero when a call fails.
Reserve, settle and release
Before a metered call starts, the system reserves its estimated usage from the account credit pool so concurrent calls cannot spend the same balance. When the call ends, it settles against actual token and media usage: unused reserved credits are returned, a failed call releases its reservation, and a successful call that exceeds its reservation records the difference. Usage reports count successful settled calls; in-progress reservations are shown separately.
All workspaces, members, Portal clients and approved follow-up questions under the same billing account share one credit pool. Plan credits are the allowance for the current window; purchased top-up credits are a separate account balance. Calls use available plan credits first, then top-up credits. A workspace cannot reserve or receive a separate credit allowance.
Allowances, resets and top-ups
- Free provides plan credits in a daily window; paid plans provide them in a monthly billing window. See the pricing page and in-app usage page for the current window and allowance.
- Purchased top-up credits do not expire.
- When a window ends, its daily or monthly plan counter resets; the top-up balance does not.
When both the account's plan allowance and top-up balance are insufficient, metered calls are refused with an explanation rather than silently degraded. Owners and collaborators can also set a monthly spending limit for an individual member; that is separate from the account pool.
Watching usage
The usage pages show the account pool's plan allowance, top-up balance, consumed credits and in-progress reservations. They can break successful consumption down by workspace, agent, caller and date. Blocked or failed attempts do not inflate successful usage. The account capacity card also shows workspace, storage, dataset, scheduled-job and published-Portal usage against its limits. Check it before upgrading: growth may come from a scheduled job, a chatty agent or Portal follow-up questions rather than general usage.
Keeping usage down
- Give agents narrow scope. Fewer metrics in context means fewer tokens per turn.
- Write good metric descriptions. They add a small amount of context and save whole retry loops.
- Reserve video and audio for deliverables people actually watch.
- Schedule recurring runs at the cadence you read them, not hourly by reflex.
- Disable unneeded tools or connections in workspace tool settings; tool calls use the same account credit pool.