Skip to main content
A workflow has two kinds of run. A test run tries the draft you are editing and sends nothing to other apps. A production run executes the live release you deployed, and its actions are real. Past runs labels each Test or Production.
Every button that tries your draft is called Test run, and says so: Test run in the builder header and in Past runs, Test run from here and Test run this step only on a step, Test run full flow in a form trigger’s Test tab. Run live now in the builder header starts a production run of the live release, after you confirm it.

Test runs

Every test run tries the draft as it is now. You do not need to deploy first. Start one:
  • Test run in the builder header, beside Deploy — runs the whole draft. Its tip reads “Runs your draft — nothing is sent to other apps, but AI steps use credits”. On a workflow a form starts, it opens the form’s Test tab so you can type sample answers first;
  • Test run in Past runs — the same run (“Uses your draft — nothing is sent to other apps”);
  • Test run from here (a step card’s play button) — from that step onward; on the trigger it runs the whole workflow;
  • Test run this step only, from the card’s run menu — just that step.
The builder saves pending edits first. If they cannot be saved, the run does not start, so a test never runs an older copy than the one on screen. Once the run starts, Past runs opens on it. If a test run cannot start because a step is not set up correctly, Hive refuses it, names that step and offers Show step to open it, the same way it does for a refused deploy. The builder’s next step suggests a test run on a draft that has never had one: Run a test, or Try the form when a form starts the workflow. It reminds you what a test does and does not do: “Nothing is sent to other apps, but AI steps use credits and any PDFs or images are saved to the Library.” A test run is safe to repeat:
  • every action is shadowed: Hive records what it would have done but sends nothing and requests no approval. Past runs shows the step as “Not sent: this was a test run” and says what it would have done in a line: for a message, the recipient and subject, for example “Would have sent email to sam@example.com: Your invoice”; for other actions, a short form of what it would have sent. The full payload is in a Would have sent box under Show raw output;
  • each send’s recipients are checked as a production run would check them: an email address that is not valid, a Slack channel the connection cannot see or a Slack connection that is missing a permission fails the step with the reason, and the test stops there. The step’s detail starts Test run: not sent., the toast shows the reason, and Past runs opens on the run. See Test runs check recipients;
  • an Agent library step runs its pinned agent revision with your own access, but takes none of the agent’s selected actions: the agent says what it would do instead, and Past runs notes “Test run: the agent took no actions; it says what it would do.” A production run’s agent step takes them as usual;
  • a timed wait is not waited out; the step notes that the wait was not observed;
  • test runs never send failure notifications.
Agent, Agent library, AI condition, web search and image steps still call their models in a test run, and use credits like a production run: your own AI credits on Business, the workspace’s pool on an earlier plan. Generate image and Create PDF steps save their files, marked as from a test run.
That spend appears in the workflow’s Model cost on its All workflows row. Runs counts test runs separately from production runs, and the row’s run date includes them — see Runs and cost in the list. A test run uses your own connected accounts and needs the Operator role or higher. When you test a workflow someone else owns, its Slack steps are checked against your Slack connection, while its production runs use the workflow owner’s, and a failed step says so. Each test run, including Test run this step only, counts toward the workflow’s Monthly credit limit and Per-run credit limit, exactly as a production run does, and draws on your credits (on an earlier plan, the pool) rather than test credits. It does not count as one of your workflow runs. The workspace’s usage lists test runs as Workflow tests, apart from live Workflows. See Credit limits.

Test a form trigger

A form trigger has a Test tab. Choose Trigger only or Full flow, then:
  • Fill form — type sample values for the draft form’s fields; or
  • Replay events — resend one of the last 20 real submissions exactly as recorded. If the form has changed since, Hive names the field that no longer fits.
Then select Test run full flow or Test run trigger only. Sample input is validated like a real submission, for example “Email is required by the trigger form.” A manual or scheduled trigger takes no sample input.

Test a connected app event trigger

A Connected app event trigger has a Test tab too, so you can test a draft before any real event arrives. Choose Trigger only or Full flow, then:
  • Sample event — edit the event as JSON. It starts as a realistic example of the trigger’s event: for an Email received trigger on Outlook or Gmail, a whole email with its subject, fromName, fromEmail and body; for other apps, the event type alone, to which you add the fields your steps read. Test run stays unavailable until the text is a JSON object; or
  • Replay events — resend one of the last 20 real events that started the deployed workflow, exactly as the app sent it.
The run records the event the way a real one is recorded, so a step reads it as {{trigger.event.<field>}}, for example {{trigger.event.subject}}. The event’s provider is the trigger’s app. A sample the trigger would not start on is refused before the run, naming what the trigger listens for, for example “its trigger listens for Email received”. Test run in the header and Past runs use the sample the Test tab holds. No mailbox holds the sample email, so a Read from an app or Agent step that opens that message or its thread reads the sample itself, and the step says it used the sample email. A live run reads the real message.

Deploy a release

Deploying turns the draft into a release (also called a version): a frozen, numbered copy of the workflow that production runs use. Deploying makes it live. You need Can edit on the workflow to deploy it, and your role must allow every action step in it.
1

Open Deploy

Select Deploy in the builder header. The dialog reads “Make your current draft the version that runs for real.”
2

Fix anything under Before you deploy

If anything would stop the deploy, Before you deploy lists it, each problem naming its step with Open step to go straight to it. Until they are fixed, Confirm deploy is unavailable, and its tip says how many are left, for example “Fix the 2 items above first”. There is no way to deploy anyway. Steps that still need an account are listed too, but do not hold the button. Below the list, the dialog says if you have never tested the workflow (“You haven’t tested this workflow yet. Do a test run first.”) or your latest test run failed (“Your last test run failed at “Draft reply”.”). That is advice only: it never stops the deploy. See Whether you tested it.
3

Review the changes

Changes lists steps added (+), changed (±) or removed (−) and connections rewired since the live release. Editing a comment, a title or a card’s position is not a change. Any warnings sit below; the list opens by itself when there are one or two. Select a changed step to see what changed in its settings. See What changed in a step.
4

Describe and confirm

Optionally enter a Deploy description (up to 500 characters), for example “Added an approval step”, then select Confirm deploy. Left empty, a first deploy is recorded as “First version” and a later one as a summary of what changed, such as “Added 1 step, changed 2 steps”.
New version shows the number the release gets. Patch is chosen for you. To pick another, open Version number: Patch (Small fix), Minor (New capability) or Major (Big change). A first deploy is 0.0.1, 0.1.0 or 1.0.0, and each later deploy bumps the most recent version. The number on the header’s Deploy button counts the changes not live yet (“3 changes not live yet”); it is hidden before the first deploy. The list under Before you deploy names each problem in plain words, for example:
  • a step that needs an AI model the workspace does not have: ““Draft reply” needs an AI model. Ask an admin to choose one in AI settings.”;
  • a step that needs an app account connected or chosen: “Choose which Xero account “Find overdue invoices” should use.”;
  • an app step with required fields left empty, or a step that reads a value from a step that does not run before it: ““Send reminder” uses a value from a step that doesn’t run before it — pick another value.”;
  • a Connected app event trigger listening to an app you have not connected.
That last one is a warning on the trigger, not a problem that stops the deploy: the workflow deploys, but its trigger never fires until you connect the app. If Hive cannot check your connected accounts at that moment, it says so as a warning too, and Confirm deploy stays available. Steps you switched off are not problems: a warning says how many will be skipped, for example “2 inactive steps will be skipped”. The exception is a switched-off Condition with switched-on steps after it, which the deploy refuses (see below). A workflow that runs on its own runs as a person: each saved schedule as the member who created it, and a Schedule trigger with no saved schedule, a public form or a Connected app event trigger as the workflow’s owner. When a Create PDF step saves to a Library folder one of those people can’t write to, the deploy preview warns on that step, naming the person, for example “‘File recap’ saves to a folder Sam Patel can’t write to. Choose a folder they can write to.” A folder that was deleted gets the same warning. Each person is checked with their current role, so the warning appears even when you can write to the folder yourself. The warning does not name someone whose runs are refused whatever the folder: someone who no longer holds the Operator role or above, for example someone listed with No role on Members, or someone with no Basic, Pro or Max seat while seats are enforced. Their runs stop before any step, so changing the folder would not help. If you can’t use the Library yourself, the deploy preview leaves the warning out. It is a warning, not a problem that stops the deploy: Confirm deploy stays available, and the runs that try to save there fail at that step. To fix it, choose a folder everyone the workflow runs as can write to, or leave the step’s folder empty to save to the Library root. Hive says nothing when it cannot check a folder at that moment. An action step that sends somewhere the workflow does not trust to choose (see When an action waits for approval) gets a warning too, naming where that comes from: “This step will wait for approval each time it runs: where it sends comes from a public form.” Inside a loop it reads “This step will wait for approval once for each item: where it sends comes from a record a Read from an app step read.” A step that writes to a connected app is warned only about the values it writes: “This step waits for approval when an address, link or number in what it writes comes from a public form.” The deploy goes ahead; to have the step send without waiting, write where it sends into the step itself. A Web search or Generate image step on Automatic whose only run data would come from steps that read your connected apps is warned about as well, because an automatic query or prompt leaves that out (see Web search and Generate image). Hive checks the whole workflow again when you confirm, and refuses, naming the step, when:
  • the graph is invalid, has a cycle, or a branch leaves a loop;
  • a Condition is switched off while steps after it are still switched on, because every path after it would run;
  • a loop or wait has no limit, or the workflow could exceed the 500-step run budget;
  • a Wait step is set to Until a value appears, which can never be met; switch it to a fixed duration;
  • a step reads a value that does not exist (in a placeholder, or in a Condition’s comparison value), reads a step set to continue after a failure, or reads a loop’s result from inside that loop;
  • an action is above your role, or a connected-app step is missing a required field or account;
  • a connected-app step makes a call Hive cannot run, such as an app address Hive has no reviewed action for (see A step Hive cannot run);
  • a step needs something your workspace does not have: an AI model, web search, image generation, Library file storage or Agent library agents;
  • an AI condition is blank, or a fixed web search contains a placeholder;
  • deploying would take you past your plan’s limit on live workflows (5 on Starter; Business, Team and Enterprise have no limit): “Your plan allows 5 active workflows. An owner or admin can upgrade in Billing.” A workflow that is already live does not count again when you redeploy it, and a paused one does not count at all;
  • the trigger is a Connected app event and your plan does not include app events (they need Business or Team, and the 7-day trial does not include them): “Your plan does not include app event triggers. An owner or admin can upgrade in Billing.”
If a deploy is refused anyway, the message stays in a row under the builder header in plain words, with Show step to open the step it names, until you dismiss it. Fix that problem and deploy again; Deploy is refused lists the messages and fixes. Each deploy is recorded in the audit trail. After deploying, keep editing freely: the live release runs unchanged until you deploy again. The refusal, and each problem in the Before you deploy list, appears in the language you use Hive in, naming the step and the fix. A problem that depends on a limit, a name or a setting you chose keeps its specific wording with those details, for example:
  • too many steps, or a workflow that could run more steps than allowed;
  • a loop or wait outside its limit;
  • too many Agent library agents in one run;
  • a retry count above what the workspace allows;
  • an output key another step already uses;
  • a step reading a value another step does not produce;
  • a tool, model or app action this workspace cannot use.
A deploy refused because a step’s app is not connected asks you to connect that app. If the account the step uses is gone, it asks you to reconnect or reselect it. If two deploys of the same workflow start at the same moment (a double-click, or a deploy from an AI client while you deploy in the builder), the first one goes live and the second is refused with “Version … was just deployed in another session”. Nothing is lost: reopen Deploy to see the release that went live.

Whether you tested it

The Deploy dialog looks at your own test runs of the workflow, the ones listed in Past runs: It is advice: Confirm deploy stays available, and the dialog offers no button for it. Start a test from the header’s Test run, then come back to Deploy. A teammate’s test runs do not count, so it never says “you” about someone else’s run. The dialog says nothing rather than guess while Past runs is still loading, when your latest test is older than the runs it has loaded, and when only a teammate has tested the workflow.

What changed in a step

Select a changed step under Changes to see what changed in its settings since the live release. The card shows only the lines that changed, with one unchanged line above and below each for context; two or more unchanged lines in a row fold into one line such as “4 unchanged lines”. Where a card sits on the canvas and Hive’s internal id for the step are left out, since neither changes what the step does. A step whose only change is the account it uses still shows that change.

Versions and rollback

Past versions, in the Deploy dialog, lists every release. Select Re-activate on an earlier one to make it live again. Nothing is deleted, and the next deploy still numbers from the most recent version. Changing the live version ends the current public form link (create a new one), and can pause a schedule someone else owns until they review it. See Schedules and public forms.

Pause and resume

Pause a live workflow to stop every production run without deleting anything. Select Pause in the builder header. The confirmation names the workflow; select Confirm to pause it, or Keep running to back out. Pause and Resume appear in the header only once a workflow is live or paused. You can also pause from Deploy: its Live status section, shown once the workflow has been deployed, has Pause workflow and the same confirmation. While a workflow is paused, no production run starts from any source: not a schedule, a form, Run live now, a teammate’s shared run or an event from a connected app. Runs already in flight are not cancelled, and test runs still work. To resume, select Resume in the header, or open Deploy and select Resume on followed by the version that was live; that re-activates the release.

Production runs

Production runs execute the live release, never your later draft edits. They start from:
  • Run live now in the builder header — see Run live now;
  • a schedule, or Run now on a schedule;
  • a public form submission, or Fill in form for an in-app form;
  • an event from the owner’s connected app, for a Connected app event trigger;
  • Run with my accounts by a teammate you shared the workflow with.

Run live now

Run live now appears in the builder header once a workflow is live (deployed and not paused) and you are allowed to run it. It is never offered on a draft, nor while previewing another role. What it does depends on the trigger:
  • Manual or Schedule — starts a production run of the live release straight away, the same run a schedule’s Run now starts;
  • In-app form — opens the live form (the same dialog as Fill in form); submitting it is the run, and the dialog says it is a live run;
  • Public form and Connected app event — not offered: a public form is filled in through its link, and an event trigger needs the event.
Before a production run starts from Run live now, Run with my accounts or a schedule’s Run now, Hive asks you to confirm it. The dialog lists each action the live release takes in another app, by step title and app — for example “Post to #hive-workflow-test — Slack”. Select Run live to start it, or Cancel to start nothing. Run live stays unavailable until the list has loaded. If the list cannot be read, the dialog says so with Try again, and the button reads Run live anyway: the run still takes its actions as usual. Steps that stay inside Hive (internal notifications, drafts and documents) and switched-off steps are not listed. Whose connected accounts a production run uses depends on how it started: A step can only ever use the running person’s own account — never a teammate’s, whatever their role. Each action is carried out on the account selected for it and follows that account’s level and any workspace ceiling: on Read it is not run, and on Draft it waits for approval. On Act it runs, unless where it sends came from a public form, an app event, an app’s records, an AI step, a web search or a run an AI client started: then it waits for approval as a card, and the run carries on (see When an action waits for approval and What agents and workflows do by themselves). A run waits for a person at an Approval step you add. A scheduled run, or one started by a connected app event, also pauses before its 51st agent step until an Owner or Admin approves it from the card in their Inbox. If nobody decides within 72 hours, the run stops as Expired (see Runs past 50 agent steps). An action your workspace’s safety rails stop — an email to an address that is not valid, a direct message to someone who is not a workspace member, a post to a channel your chat tool does not reach, the hourly action limit, a pause after three failures on that account, or a spend cap — is recorded on its step as not sent, with the reason. When the kill switch, the workspace-wide emergency stop, is engaged, actions are not dispatched, unattended runs are refused, and a run already in progress stops before its next model call. A saved agent step already running stops before it acts, reaches outside Hive or saves its result; until then it may make further model calls, which are charged as usual. Production runs, like test runs, use AI credits (on Business, those of the member the run is charged to; on an earlier plan, the workspace’s pool) and count toward the workflow’s credit limits. On Business, each production run also counts as one workflow run.

Failure notifications

When a production run fails, the person it ran as gets a card in their notification bell. For a run that started on its own that is the schedule’s owner, or the workflow’s owner for a public form or a connected app event, so a workflow nobody is watching still says when it stopped working. You do not have to set anything up for this. Addresses already listed under Notification recipients in the workflow’s settings are told as well, if they belong to a member of this workspace who can open the workflow. Each address is matched to a member; one that matches no member is marked “Not a workspace member — won’t be notified”. People who can edit the workflow can remove a listed address. New addresses can no longer be added there: choosing members to notify by email is coming. Both kinds of card cover a scheduled run or a public form submission Hive refused to start, for example because the kill switch is on or the workflow reached its Monthly credit limit.
  • Only members of this workspace who hold a Basic, Pro or Max seat are notified, and only those who can open the workflow see the card. Hive sends no email.
  • Test runs never notify.
  • Each person gets at most one card per workflow per clock hour (UTC), however many runs fail in it. Every failure is still in Past runs. Someone who is both the run’s owner and on the recipient list is told once, not twice.
  • The card names the workflow and how the run started, never the error. Selecting it opens the run that failed.
  • A run that expires or is rejected at an Approval step gets its own card. See Who is told when a run expires or is rejected.
The workflow’s Monthly credit limit and Per-run credit limit, under Limits in its settings, apply to every run, test runs included. Runs per hour counts only production runs; test runs never count toward it (see Credit limits). With Usage alerts on, the workspace’s Owners and Admins and the workflow’s owner are alerted when a deployed workflow’s runs reach 50%, 80% and 100% of its monthly credit limit; the workspace’s own usage alerts cover its credits either way.

Runs and errors

Follow a run, retries and cancelling.

Schedules and public forms

Start production runs automatically.

Review and approve

Approval cards, including the workflow actions Hive holds.

Triggers and steps

The complete step reference.