Plan limits
On Business and Enterprise, AI credits and workflow runs are the only plan limits. There is no plan limit on active workflows, hourly schedules, custom agents, public agent links and public forms, connected AI clients, connected accounts, file storage, the search index, images, deep research modes (exhaustive research included), a workflow’s or agent’s monthly AI spend, or answers in Slack and Teams. The limits on the rest of this page are Hive’s own protections, the same for every workspace: rate limits, request sizes, and safety ceilings that stop runaway work.
An Enterprise contract can set its own credits and runs. Owners and Admins see the workspace’s seats
and its workflow runs this month under Plan limits in Settings → Usage. A workspace on
Starter, Team or the earlier Business plan keeps that plan’s limits: see
Earlier plans.
Approver seats, and the 10 questions a month an Approver can ask
about items routed to them, are sold only on earlier plans: see
The Approver seat.
Credits and spend controls
Workflow graph
Steps
The Agent library rows are Hive’s limits for Agent library steps, and they apply to every
workspace. The Agent library agents in one run can make at most 40 model calls between them, and
each uses at least one. Agents working at once and starts per hour count per run; only the same
agent running at once (4) counts across runs. A deploy is refused when a workflow’s Agent library
steps, loops or Parallel agents steps could need more agents working at once, more starts or
more model calls in one run than these allow; see
Agent library steps in loops and parallel steps.
A step the limits refuse while it runs, such as an agent already running 4 times, fails with the
reason, see Common failure messages.
A run keeps the limits it started with: if Hive’s limits for Agent library steps change while a
workflow is running, its later Agent library steps are not affected. See
Agent library steps that start more than 30 days after the run’s first one.
Runs
Forms
Workflow assistant
Ask Hive
Library
Larger uploads are an optional workspace capability. When enabled, the Upload files dialog
shows the applicable format limit and waits for the worker to validate a directly uploaded
file before listing it. A 256,000,000-byte file fits an empty 250 MiB standard folder, but
other files in the same folder may use its remaining capacity. Text and image uploads retain
the standard limit; enabling larger uploads does not lower their existing allowance. The
25 GiB budget also applies to standard uploads, file edits and template copies.
It is an operational safety bound, not a pricing-plan entitlement or a retroactive
limit on existing files; worker publication does not count an upload twice. See
Library.
The character sizes of file text above are for English and other text in the Latin alphabet. Hive
sizes what it reads by tokens, not characters, so text in Chinese or Japanese (about one token a
character) gets about a quarter as many characters, and text in Korean, Arabic, Hebrew, Hindi or
Thai (about one token per two characters) about half as many. The same applies to the 24,000
characters Ask reads from one web page, the open file Ask reads before editing it, and the
earlier questions it sends back with a new one.
Agent Studio
Public agent links
See Public agent links.
Hive MCP server
A member without a Basic, Pro or Max seat, including a member on an Approver seat, cannot connect AI clients, and the clients they connected earlier are refused. See Hive MCP server.
Slack and Teams
See Slack and Teams.
Signals and approvals
Signals fair use
On Business, Signals runs on every seat and uses the AI credits of the member whose accounts it reads, with no fair-use allowance. On earlier plans, Signals runs for each member with a Max seat, within these limits. See Fair use and credits.Integrations
Provider read limits
Hive limits how many accounts it reads from each provider at the same time, across all workspaces, so one large workspace or a busy provider cannot slow everyone else’s reads. One workspace uses at most half of a provider’s slots, rounded up.
These limits apply to scheduled reads: the hourly collection, file sync and the Signals read of your finance and
CRM tools. A read that finds no free slot waits for the next hourly pass; see
Reading accounts on the hourly pass.
Some quotas are shared by every Hive workspace: Hive’s connection service, Companies House, file reading,
knowledge search, Hive’s Gmail project, and each Xero organisation. Hive keeps each workspace to a fair share of
them, at most 10% of each a minute, counted in reads: an AI client’s call that reaches outside Hive is one read,
and so is one scheduled read of one connected account, even though that read can make several calls to the
provider. Scheduled reads and AI clients’ reads count together, but scheduled reads use at most half of the
workspace’s share, so the hourly pass never uses up the share your AI clients need. A scheduled read past its
half waits its turn for the share to refill and then reads, in the same pass: a pass with more accounts than the
share allows a minute reads them a few seconds apart, carrying on where it stopped every few minutes. A read that
cannot get its turn within the time it can wait, or that comes while Hive cannot count the share, moves to a
later pass; a read that does not run does not count against the share. An AI client’s read past the share is
refused with a message to try again in a minute. Each connected Xero organisation has its own share.
Signals passes
Related
Runs and errors
How the 500-step budget is counted.
Public form API
Rate limits and errors in context.
Billing and plans
Plans, seats, credits and top-ups.
Triggers and steps
Every step’s settings.