configure
- open settings (
⌘,) - go to the webhooks section
- paste a url into webhook url
- leave it empty to disable
events
request
- method:
POST - headers:
Content-Type: application/json,X-Miniti-Delivery-Id,X-Miniti-Attempt - timeout: 15 seconds for queued deliveries
- retries: network failures, 429 and 5xx responses are retried. see when a delivery fails
- transport: https only (v2.8.0). a plain
http://address is refused, in settings and at send time, because the payload carries the full transcript and attendee names. the one exception ishttp://localhostor127.0.0.1for a hook running on the same machine - signing: none (the webhook url is the shared secret - keep it private). miniti never writes the url to its debug log, only the host, since zapier, make, and n8n put the secret in the path
payload
field notes
speaker_emails(from v3.2.0) maps a transcript speaker id to a calendar attendee’s email, and each attendee carries thespeakerid it was matched to. a speaker gets an email when its name matches one attendee cleanly (full name, a unique first name, or the email’s local part) or when you pick the attendee in the rename sheet; nothing is guessed, so some named speakers have no emailmeddpicc,training,questions,attendees, andspeaker_namesare omitted when empty - do not assume these keys are always presentspeaker_namesis a map from the underlying diarization key to the resolved real name (e.g."jane doe"). the key is whatever miniti uses internally to identify that speaker - it is not guaranteed to be a human-readable"speaker N"string, so treat it as an opaque id and look up names by key rather than parsing it. see speaker names for how miniti resolves them.transcript[].speakeris the display label for that segment: the resolved real name when one is available, otherwise the display form of the diarization label, or"you"for your own mic. do not assume it matches thespeaker_nameskey.calendar_event_idis present only if the meeting was linked to a google calendar eventtranscript[].timestampis seconds from the start of the recordingtranscript_edited_atis present after a saved transcript has been trimmedtranscript_revisionincrements each time the saved transcript is editedinsights_staleis true when transcript edits have cleared generated insights and the meeting has not been regenerated yet- after transcript trimming, generated insight fields may be empty until the user regenerates insights
dateandend_timeare ISO 8601 utc- keys are always json-sorted - downstream consumers can rely on stable ordering
matching records in other systems
- use
meeting.idas the external record id for the meeting. deduplicate delivery attempts separately withX-Miniti-Delivery-Id.meeting.updatedkeeps the same id, so downstream systems can replace the existing record instead of creating a duplicate. - if your system groups meetings by customer or company, match against
meeting.attendees[].domain. this is available when the meeting came from google calendar attendees. - use
meeting.calendar_event_idwhen you need to link back to the source google calendar event. - action items do not have their own ids today. if your downstream system needs one row per action item, derive a stable id from
meeting.idplus the action item index or a hash of the action item text.
receiver requirements
- respond with any 2xx status only after you have saved the payload or durably queued it
- respond within 15 seconds or the queued delivery times out and may retry
- handle repeated delivery ids without repeating side effects. delivery attempts may repeat; the finite retry window does not guarantee receipt
- accept a payload up to ~a few mb for long meetings (transcript size dominates)
recipes
- zapier / make / n8n - use their built-in “webhook by url” trigger and map fields from the
meetingobject - slack - forward summary + action items to a channel by transforming the payload in a small middle function
- notion / airtable - create a row per
meeting.savedevent, mappingid,title,summary,action_items
see also
- attio integration - native push for attio users
- Twenty CRM integration - native push to Twenty people, companies, and opportunities
- markdown export - file-based alternative that works offline
when a delivery fails
from v2.11.0 a webhook that cannot be delivered is kept and retried: after 30 seconds, then 2, 10 and 30 minutes, then 2 hours, for up to a day. a 4xx response other than 429 is treated as final and not retried. settings → webhooks shows how many deliveries are waiting or failed, with retry now and discard failed. every request carries anX-Miniti-Delivery-Id header (stable across retries of the same delivery) and X-Miniti-Attempt, so a receiver can ignore repeats if a response was lost after it had already handled the payload. through v3.5.0, an updated-meeting delivery that is still waiting can be replaced by a newer version rather than queued twice. the 3.6.0 changes below give attempted deliveries their own immutable body. deleting a meeting drops its pending deliveries.
changes in 3.6.0
available in macos and ios 3.6.0. older versions may send
meeting.saved before final ai processing finishes.meeting.saved is queued after the stopped meeting and its final processing outcome are saved, even if you leave that meeting open or go home. the payload contains the saved transcript and available summary. it does not wait for cloud sync, need pro, or require the receiver to fetch the meeting from the cloud.
when processing fails, is skipped, or is interrupted, the transcript can still be delivered with the results available so far. an optional top-level processing object explains that outcome:
processing.state is complete, partial, failed, skipped or interrupted. each entry in processing.outcomes is complete, failed, skipped, interrupted or stale. keys identify processing modes, such as standard for the summary. the map can be empty when a run was interrupted before its per-mode results were saved. accept a missing processing object from older clients and do not assume a summary exists just because a webhook arrived.
once a delivery has been attempted, its id and body stay together for retries. newer meeting content gets a new delivery id; finishing an older attempt cannot discard the newer delivery. retries of superseded content stop. this does not guarantee ordering across devices.
retry now retries a waiting delivery’s saved body. resend current meeting creates a new meeting.updated delivery from the meeting as it is saved now, including transcript edits and the available summary. it gets a new delivery id. old failed payloads are not resurrected; deleted meetings cannot be resent.
changing accounts does not cancel an already configured webhook export. ai results from the previous account are still rejected. if you change the webhook url while final processing or the saved-meeting handoff is waiting, miniti asks you to try again before sending that waiting completion to the new url. deliveries already released for sending keep the existing behavior of using the currently configured url. later updates can go to that configured url while the initial completion waits for your decision; receivers should accept meeting.updated even if they have not received meeting.saved.
while final ai is running, update shows finishing in the stopped meeting and history. wait for it to finish before requesting another ai update. coaching still works locally. changing the selected template does not discard the summary or questions.
staged deliveries also expire after 24 hours, measured from the original completion intent. expiry is applied when the app can run; it does not guarantee background execution on ios. miniti removes the queued payload and shows delivery expired with resend current meeting. changing the destination does not restart that clock. a held initial delivery does not block later updates, while an initial delivery already released for sending keeps its normal place ahead of updates.
