> ## Documentation Index
> Fetch the complete documentation index at: https://docs.get-hive.ai/llms.txt
> Use this file to discover all available pages before exploring further.

> ## Agent Instructions
> Hive is an AI operations layer for small and mid-sized business teams. It reads the tools a team already runs, raises what needs attention, and acts with the approvals the team configures.
> Answer only from these pages. If a page does not cover something, say so and link the closest page rather than inferring behaviour.
> The supported programmatic entry points are incoming webhooks, public workflow forms and the remote MCP server, each documented under Developers.
> Each connected account has a level: Read, Draft or Act. New accounts start on Act. Agents and workflows take only the actions selected for them, whatever the account level.
> Describe security, sign-on and data handling only as the Security pages and the FAQ describe them, and add no certification, compliance, single sign-on or model-training claim they do not make.
> Prices and credit allowances are in GBP. Quote them only as a page states them, without converting.
> Use UK English and the product's exact UI labels. Link to pages as https://docs.get-hive.ai/<path>.

# Build workflows, schedules and agents from Ask

> Ask Hive to schedule a prompt, set a reminder, draft a workflow or agent, or run, change or delete a workflow. Nothing happens until you press the card.

Ask can set things up for you as well as answer. Say what you want in your own words, and Hive
shows a card with exactly what it would create. Nothing is created, scheduled or run until you
press the card's button. This is the **Workflow & schedule builder**.

It works on Hive's newer Ask engine, in a saved chat on the web: your own, or one shared with you
as **Can collaborate** (see [In a shared chat](#in-a-shared-chat)). Slack and Teams answer only.

## What you can ask for

| You say | Hive shows |
| - | - |
| "Every Monday at 8, send me a summary of overdue invoices" | **New scheduled task** |
| "Remind me on 14 Oct at 14:45 to call Sam back" | **New one-time task** |
| "When a new HubSpot deal is created, research the company and post to Slack" | **New workflow** |
| "Make an agent that summarises my unread Outlook email each morning" | **New agent** |
| "Run my Weekly status workflow now" | **Run Weekly status** |
| "Tell me when an email from Vanta arrives" | **Follow this as a matter** |
| "Every time HMRC emails, summarise and file it" | **New workflow** |
| "Add this to the Plant project" | **Add to Plant** |
| "Add a step to Deal research that posts the summary to Slack" | **Change Deal research** |
| "Delete my Old intake workflow" | **Delete Old intake** |

A "keep an eye on…", "tell me when…" or "let me know if…" request is either one thing to follow or
something repeating, read from your words. One thing ("tell me when", "as soon as", "keep an eye
on") is a **Follow this as a matter** card: Hive follows it as one of your
[matters](https://docs.get-hive.ai/signals/matters.md#ask-hive-to-keep-an-eye-on-something). Something repeating ("every time",
"whenever", "from now on") is a **New workflow** card that starts on the app event, marked
**Every time** with your words.

## The card

Every card starts as a draft and has the same buttons:

* **Create** (or **Run**, **Add**, **Save** or **Delete**) does exactly what the card shows, as
  you.
* **Edit in Workflows** or **Edit in Agent Studio** creates the draft and opens it in its builder.
* **Cancel** closes the card. Nothing is created.

The card says what each draft contains:

> [accordion] New scheduled task
>
> What runs, either a prompt or one of your saved agents. When it runs, in plain words with your
> timezone ("Mondays 8:00 AM, London"), and the date of the next run. Where the results go, and
> the model a prompt runs on.

> [accordion] New one-time task
>
> The same card for something that runs once: when it runs, in plain words with your timezone
> ("Once on Wed 14 Oct 2026 at 2:45 PM, London"), under **Runs at**. After it has run it is
> finished, and it never runs again.

> [accordion] New workflow
>
> The workflow's name, its trigger in plain words, its numbered steps and the apps it uses. Each
> app shows your **Read**, **Draft** or **Act** setting for it. That setting governs only Ask and
> Signals: the workflow's action steps run whatever it is set to.

> [accordion] New agent
>
> The agent's name and description, and the first lines of its instructions (select **Show all
> instructions** for the rest). Also the model, the connected apps with the actions it may take,
> any Library folders from this chat, and its skills.

> [accordion] Follow this as a matter
>
> What Hive is waiting for, in your words, and the accounts Signals reads for you, so you know
> where it will look. When one of your own open matters clearly fits (the same invoice, deal,
> person or thread), the card is titled **Add to** and that matter, and adds **Waiting for: …**
> (set by you) to it; otherwise it starts a new matter of yours, titled as shown. Press **Follow**
> (or **Add to matter**); **Open matter** then opens it on the Signals **Matters** tab. In a shared
> chat, only the person who asked can press it. See
> [Ask Hive to keep an eye on something](https://docs.get-hive.ai/signals/matters.md#ask-hive-to-keep-an-eye-on-something).

> [accordion] Run a workflow
>
> The workflow's form fields, filled in with what you said: the form of its deployed release,
> which is what the run uses, even when a change to the form is saved but not yet deployed.
> Hive does not fill in a field that decides where a step sends, such as an email address, a
> channel or a phone number, from what you said: type it on the card yourself. That includes a
> field a step that writes to a connected app uses for an address, link or number. An older run card
> that filled in such a field is refused when you select **Run**, with "This card was made before
> an update: ask Hive again."
> Change any of them on the card, then select **Run**. The card then shows the run's status as it goes, from running to done, and links
> to the run. A run Hive estimates at more than £5 (500 credits on an earlier plan), or one that
> could take more than 50 agent steps, asks first, as it does in Workflows: the card shows the
> estimate and **Run anyway (up to about £…)** (**Run anyway (up to … credits)** on an earlier
> plan), and the run is then held to that estimate. See
> [Confirming a large run](https://docs.get-hive.ai/workflows/runs-and-errors.md#confirming-a-large-run).

> [accordion] Add to a project
>
> What it adds to one of your projects: this chat, a file you sent in it, or both. It says that
> everyone with access to the project can read what you add, and how many named collaborators it
> has. After **Add**, **Open** with the project's name goes to it. See
> [Ask Hive to add it](https://docs.get-hive.ai/ask/projects.md#ask-hive-to-add-it).

> [accordion] Change a workflow
>
> The workflow's trigger and its steps in the order a run takes them, as they will be after the
> change. A step the change adds is marked **New**, one it alters **Changed** (with what the step
> now does), and the steps it takes out are listed under **Removed**. **Save** saves exactly
> these steps. If the workflow is deployed or paused and the change alters what a run does, the
> card says the change is saved to the draft only: what is deployed stays as it is, and the
> change takes effect once you deploy it in Workflows. Renaming a step alone needs no deploy. An
> older deployment with no current release goes back to a draft instead, and the card says so
> before you save (see [Draft, live and paused](https://docs.get-hive.ai/workflows/workflow-builder.md#draft-live-and-paused)).

> [accordion] Delete a workflow
>
> The workflow's status and how it starts. **Delete** does what **Delete** does in Workflows: the
> workflow leaves your workspace and its schedules stop. Its runs and versions are kept for
> audit.

After **Create**, the card becomes a confirmation with **Open**. A scheduled task also shows its
next run time, and a one-time task the time it runs. A new agent also gets **Try it**, which selects the agent in this chat, just as
picking it from the **@** menu does.

## Change a draft by chatting

Reply to change the draft: "make it 9am", "also post to Teams", "use GPT-5.5". Hive updates the
same card where it first appeared, and your reply shows a note pointing to it. Once you have
created or cancelled a card, a new request starts a new draft.

## Work on your existing workflows

Ask can find, open, change, run and delete the workflows you can see in Workflows.

* **Find and open.** Ask "which workflows do I have?" or "what does Deal research do?". Hive reads
  your workflows and answers straight away: each one's status, how it starts, its steps in run
  order, its apps, and whether you can change, run or delete it. Reading needs no card. For a
  live or paused workflow it describes the deployed release, which is what runs, and says when a
  change you saved is waiting for the next deploy.
  * **Schedule.** A scheduled workflow's start gives its schedule in the time zone it runs in, and
    the date and time of its next run there, for example "Mondays 9:00 AM (Europe/London; next
    run Mon, 2 Nov 2026, 9:00 am)". If the time zone you are using Hive in is different, Hive also
    gives that next run in your time. A schedule that is switched off is said to be off.
  * **Whose schedules.** A workspace Owner or Admin sees every schedule of a workflow. Anyone else
    sees only the schedules they set up, and Hive says so. An older workflow whose schedule is
    kept only in its start step runs in UTC; Hive says UTC only when it can see that the workflow
    has no other schedule. Otherwise it gives the start step's time and says the time zone is not
    shown, rather than guessing one.
  * **Agent steps.** For a step that runs an Agent Studio agent, Hive names the agent and reads
    the instructions of the version the workflow runs, only if you can open that agent in Agent
    Studio: you own it, it is shared with you, or you are a workspace Owner or Admin. Otherwise
    Hive says its instructions are not available to you and does not name the agent.
  * **Long workflows.** Hive describes up to 30 steps of a workflow and opens up to 10 agents, and
    shortens long instructions. It says how many more steps there are; open the workflow in
    Workflows to see them all.
* **Change.** Say what to change ("add a step that posts the summary to Slack", "remove the email
  step"). The [workflow assistant](https://docs.get-hive.ai/workflows/workflow-assistant.md) drafts the change over the
  workflow as it is, and Hive shows a **Change** card. Reply to refine it before you press
  **Save**. Ask for changes in your own chat: a chat someone shared with you offers no
  **Change** card. To rename a workflow or change who it is shared with, use Workflows.
* **Run.** Ask "run my Weekly status workflow now" for a **Run** card, as above.
* **Delete.** Hive shows a **Delete** card for one workflow at a time. To delete several, ask Hive
  to list them and confirm each card.

Changes are saved only if the workflow has not changed since the card was drafted. If someone
changed it in Workflows in the meantime, the card says so and nothing is saved or deleted: ask
again to draft it over the current version.

## How scheduled tasks work

* **When.** Say how often and at what time: hourly, daily, on chosen weekdays ("every weekday at
  9:30") or, for an agent, on a day of the month ("on the 1st of every month at 9am"). The time is
  read in your own timezone, the one Hive stored from your browser the first time you opened
  the app, or London if you have none. Hive cannot run an interval such as "every two weeks", and
  tells you so in a sentence.
* **Once.** Say a date and time, and the task runs once then: "14 Oct at 14:45", "on 14/10 at 9"
  (day first), "tomorrow at 9am", "first thing tomorrow" (9:00 AM), "Friday at 10" (the coming
  Friday), "in 2 hours". A date with no year is the next one to come. Say a time as well as a day;
  Hive asks for one when you do not. The time must be in the future and at most a year ahead, and
  is checked again when you press **Create**. A time zone you name in the text ("BST") is not
  read: the card shows the timezone Hive used. After it runs, the task is finished and shows
  **Finished** on the **Scheduled** page; turning it off and on again does not run it again. Hive
  never sets up a repeating schedule in place of a one-time task.
* **A prompt** becomes a small workflow with one AI step, on the model you chose for this chat. It
  is deployed and scheduled when you press **Create**. Each run's answer is saved in the
  workflow's **Past runs**. A failed run notifies you in your Inbox. When a one-time task has run,
  your Inbox tells you ("One-time task ran: …"), with a link to the workflow; the answer itself
  stays in **Past runs**. This also happens when the task waited for credits and ran later.
* **An agent** must be deployed first. The schedule runs its deployed version, and each run is
  saved with the agent's runs in Agent Studio. If Hive does not run a one-time agent task (for
  example, the agent was already running or has changed, or Hive was unavailable at the time), your
  Inbox tells you ("One-time task didn’t run: …"), with a link to the agent. The task's receipt on
  the **Scheduled** page says why, and it does not run again. See
  [Schedule an agent](https://docs.get-hive.ai/agents/publish-and-evaluate.md#schedule-an-agent).

Every scheduled task appears on the **Scheduled** page, where you can pause or delete it. See
[Schedules and forms](https://docs.get-hive.ai/workflows/schedules-and-forms.md).

A scheduled or one-time run uses its owner's AI credits (on an earlier plan, the workspace's shared pool) and seat,
exactly as any scheduled run does: it acts as you, through your connected accounts, and a step
that needs approval still waits for it.

## Follow-ups Hive suggests

When a chat implies a follow-up you have not asked to schedule (you say "remind me…" or "first
thing tomorrow", or Hive or you promise to check back on something by a time), Hive may suggest
one: a **New one-time task** card saying what it would check and when. It is only a suggestion.
Nothing is scheduled until you press **Create**, which takes one tap; **Cancel** or ignoring it
sets nothing up. Hive suggests at most one at a time, never for a time that has passed, and not
again in a chat where you declined one. It never sets up a follow-up on its own, because a
scheduled run uses credits and can act on your connected apps.

## Workflows and agents start as drafts

* A workflow is drafted by the [workflow assistant](https://docs.get-hive.ai/workflows/workflow-assistant.md), on Hive's
  Efficient model. **Create** saves it as a private draft. Deploy it in Workflows to turn it on.
* An agent is drafted by Agent Studio's builder, on Hive's Efficient model; the agent itself keeps
  the model you chose for this chat. **Create** saves it as a draft in
  [Agent Studio](https://docs.get-hive.ai/agents/agent-studio.md). Deploying it is a separate step there.
* A **New workflow**, **New agent** or **Change** card names only the email addresses, phone
  numbers, links, channels and repositories you typed in this chat, your own connected mailbox's
  address and, on a **Change** card, those the workflow already has. When the draft needs another,
  Hive asks you to type it instead of filling it in. Nor does a card add or change an Agent step
  whose answer a step sends to, or point a send at such an answer: make that change on the canvas
  yourself.

## Save a chat as an agent

When a chat has done a job you want to repeat, choose the chat's name at the top of the chat,
then **Save as agent**. Name the agent (the chat's title is filled in) and press **Save as agent**.

* Agent Studio's builder drafts a reusable agent from what you asked in the chat, in order, and
  from Hive's latest answer as an example of the output.
* The agent keeps the model the chat last answered with, web access if the chat searched the web,
  and the Library folders and files the chat used.
* It is saved as a private draft and opens in [Agent Studio](https://docs.get-hive.ai/agents/agent-studio.md), where you can
  review it, test it and deploy it. Nothing runs and nothing is shared.
* The draft may name only the email addresses, phone numbers, links, channels and repositories you
  typed in the chat, and your own connected mailbox's address. Otherwise Hive saves nothing and
  says, for example, "This agent's draft names [finance@acme.com](mailto:finance@acme.com), which you did not type in this
  chat. Type it in the chat, or add it yourself in Agent Studio."
* You can only save your own chats, once Hive has answered at least once. You need the
  **Department lead** role or above, as for creating any agent. Drafting uses your AI credits (on an
  earlier plan, the workspace's shared pool), like an Ask turn.

## Who can do what

* Creating a scheduled task or a workflow, following something as a matter, and running a
  workflow, need the **Operator** role or above. Creating an agent needs **Department lead** or
  above, as in Agent Studio.
* Anyone who can use Workflows can ask Hive to list and open the workflows they can see.
  Changing or deleting one also needs the **Operator** role or above. You can change a workflow
  you own or one shared with you as an editor. Only a workflow's owner can delete it; Hive tells
  anyone else so.
* Hive only offers workflows you can see and have permission to run. A workflow must be deployed,
  and it must start by hand or from a form. A workflow that starts on a schedule or from an app
  event cannot be run from chat.
* A workflow's action steps run on the account selected for each step, whatever it is set to
  (**Read**, **Draft** or **Act**). Your workspace's limits still apply. A step waits for approval
  when where it sends came from a public form, an app event, an app's records, an AI step or a web
  search, and a run waits for a person at an [Approval](https://docs.get-hive.ai/workflows/workflow-steps.md#approval) step.
  The values you type on a **Run** card count as your own. See
  [What agents and workflows do by themselves](https://docs.get-hive.ai/integrations/accounts-and-autonomy.md#what-agents-and-workflows-do).
* Drafting runs on Hive's Efficient model and uses your AI credits (on an earlier plan, the
  workspace's shared pool, counting towards your own monthly credit limit if your workspace sets
  one), like any other Ask turn.

## In a shared chat

In a chat someone shared with you as **Can collaborate**, you can ask for a scheduled task, a
workflow, an agent, a watch or a workflow run just as in your own chat, with the same roles as
above.

* What you create is yours: a scheduled task, workflow, agent or watch is made in your account and
  runs as you, on your apps, seat and credits, and a workflow you run starts as you. Nothing is
  created in the chat owner's account.
* Only the person who asked can press a card. Everyone in the chat sees the card, and once it is
  used they see that it was created, but on a card from someone else's question there are no
  buttons. Hive checks again that you can still ask in the chat each time you press one.
* **Can view** members can't ask Hive in the chat, so they get no cards.
* "Add this to a project" is only offered in your own chat, because moving a chat into a project
  is its owner's choice.

## Related

> [card] [Ask Hive](https://docs.get-hive.ai/ask/ask-hive.md)
>
> Ask questions, attach files and act on your connected apps.

> [card] [Workflow assistant](https://docs.get-hive.ai/workflows/workflow-assistant.md)
>
> Draft and change workflows from a description.

> [card] [Schedules and forms](https://docs.get-hive.ai/workflows/schedules-and-forms.md)
>
> Manage every schedule on the Scheduled page.

> [card] [Agent Studio](https://docs.get-hive.ai/agents/agent-studio.md)
>
> Build, test and deploy agents.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.