Skip to main content
Hive uses a small vocabulary consistently across the product. Learn these terms once and every screen reads the same way. For a short A–Z list, see the glossary.

Who decides whether Hive acts

Each connected account has its own level, set by the member who connected it: Read (Hive only searches and summarises), Draft (Hive prepares actions and asks you first) or Act (Hive carries out actions and tells you what it did). New accounts start on Act, and an Owner or Admin can cap an app lower. This level decides what Hive does in Ask and Signals. Agents and workflows carry out the actions you select for them whatever it is set to, except one whose recipient, channel, number or link did not come from words you trust, which waits for your approval; to make a workflow wait for you every time, add an Approval step. See Accounts and autonomy.

The operating loop

Hive runs a loop over your connected tools: collect what changed, detect what needs attention, propose a next step, pass it through a deterministic policy check, execute what was approved, and learn from the outcome. The model proposes; the policy engine decides what may run. Everything is recorded in the audit log.

Signals and attention

Signals is an optional product, on by default for new workspaces. An Owner or Admin can hide it under Settings → Products → Signals. See Signals.
Something Hive found in a connected tool that is at risk of being missed — a risk, an opportunity or a decision someone needs to make. Every signal carries evidence (the readings and their source), why it matters, a severity (High, Medium, Low) and usually a proposed next step. See Signals.
Ask Hive to tell you when something happens (“tell me when the signed contract arrives”) and it follows it as one of your matters, with what you are waiting for as a requirement it checks every hour; “every time HMRC emails, summarise and file it” is a workflow. See Ask Hive to keep an eye on something.
Snooze hides a signal for a while — it comes back. Dismiss asks why, and the reason changes what Hive raises next (stay quiet about a record, mute a type, or flag a data source).

Approvals and safety

A role-gated decision on an action Hive proposes in Ask or Signals. A signal’s approvals appear on the Signals screen, with Approve & run now, Decide later and Do not run; an Ask chat’s stay in that chat, with Approve and Decline. They expire after seven days. An agent or a workflow raises one for an action whose recipient, channel, number or link did not come from words you trust, and for the few actions that always wait for you; see What agents and workflows do by themselves. A workflow also waits for a person at an Approval step you add, which asks its approvers under Needs you in their Inbox. See Review and approve.
Halts every autonomous action and approval company-wide. Owners and admins control it; it survives restarts. See Safety and control.
An immutable trace of signal → plan → decision → result for every action. Reversible actions can be undone. See Audit log.

Advanced: per-action policy

Most firms only need the account level above. Owners and Admins who want finer control set policy per type of action under Settings → Products → Governance. See Safety and control.
How far Hive may go on its own for a type of action in Ask and Signals: Shadow (simulate only), Suggest, Approve each, Auto in limits and Autonomous. There, money, irreversible and B4 actions always ask a human. The ladder works with the account’s level: in Ask and Signals, an account set to Act treats a Suggest or Approve each action as Auto in limits, and a Shadow action stays observe-only. To keep an account’s actions on approval, set it to Draft. An agent’s or a workflow’s selected actions are not on the ladder: they run by themselves, though the kill switch and the workspace’s limits can still stop one, and one whose recipient, channel, number or link did not come from words you trust waits for your approval. See Autonomy and blast radius.
How much damage an action could do if it were wrong.
Hive simulates what it would do without acting, so you can see the value before granting autonomy.

Knowledge

Each member’s own memory of their business world, built from the accounts they connect and visible only to them: entities, facts and relationships, each with its source, date and trust level. Its page in the sidebar is an optional product, on by default for new workspaces. See Business Brain.
The rules Hive learned from how you work, in Business Brain → Rules. Each member has their own, private to them, and each rule is On, Off or Suggested: Hive never turns its own rules on. See Rules Hive learned.
Notes you choose to save about yourself for chat, private to you and off by default. See Personal memory.
Your workspace’s files and folders — uploads, generated documents and images — plus a prompt library. See Library.

Doing the work

A chat that answers across your tools, files, the Brain and the web, with citations. Projects group chats with shared instructions and files. See Ask Hive and Projects.
A graph of steps started by one trigger (manual, in-app form, public form, schedule or connected app event). A live run carries out its action steps by itself, however it started, except a step whose recipient, channel, number or link came from a public form, an app event, an app’s records, an AI step or a web search: that step waits for approval. A run also waits for a person at an Approval step you add; a test run holds every action back. Workflows are drafted, tested, then deployed as numbered versions. See Workflow builder.
A specialised assistant you configure in Agent Studio for one job, with its own knowledge, connectors, selected actions, budgets and evals. Each selected action follows the level of the account it runs from (Act runs it, Draft asks you first, Read takes none), and one whose recipient, channel, number or link did not come from its instructions or your own words waits for your approval; an agent set up before actions were chosen per app takes none until you select some. See Agent Studio.
An agent run or workflow on a recurring schedule. See Scheduled tasks.

Connections

A connection to one of your tools, over managed OAuth, an MCP server or a signed incoming webhook. Each connected account is on Read, Draft or Act: see Who decides whether Hive acts.
Slack or Microsoft Teams, used to ask Hive questions from chat. Signals and approvals stay in Hive. See Slack and Teams.

People and billing

Owner, Admin, Department lead and Operator. Roles decide who can approve which blast-radius tier and who manages members, policy, connectors and billing. See Roles and permissions.
Finance, Sales, Marketing, Customer Success, People and IT & Security, plus a whole-business view. Roles can be scoped to departments.
After a 7-day free trial, Hive sells Business online and Enterprise by contract. Every member needs a paid seat to use the workspace: a Basic, Pro or Max seat. The seats have the same features and differ only in the AI credits they include each month, £5, £30 or £60. Every seat also includes 2,500 workflow runs a month. There are no free seats. Workspaces on Starter, Team or the earlier Business plan keep them: see Earlier plans. See Billing and plans.
Money for AI work: 1 credit is 1p. Each member has their own, included with their seat each month, and Hive spends them at what the work costs it, with no mark-up: a typical answer on Hive Auto costs about 4.4p. Ask, agents, workflows, Signals, the Business Brain, research and images use them. When a member’s credits are used, their AI work stops, unless an Owner or Admin has turned on extra usage: then they carry on, billed at cost to the workspace, up to their top-up limit and the workspace’s monthly maximum. On Starter, Team and the earlier Business plan, every member draws on one shared pool instead: see Earlier plans. See Credits and usage and Models.

Navigate Hive

Where each of these lives in the app.

Glossary

Every term, A to Z.