The output of a meeting is not one thing that appears and sits there. It is produced, checked, and then distributed, and a different person is responsible at each of those points. Setting out who does what makes the whole arrangement easy to run and easy to audit later.
The three states
A record moves through three states in order, and nothing skips a step.
The state is visible on the record itself, so anyone who opens it knows which of the three they are looking at.
The host, at the end of the call
When the host closes the session, the transcript and a drafted summary are produced and land in draft state. The host's work at this point is to confirm what the meeting actually produced: which decisions belong in the record and in what wording, which spoken commitments become action items, who owns each one, and which dates people named out loud. The host also names the reviewer.
- The wording of each decision, close enough to quote it back later
- An owner on every action item
- A date on the items where one was actually agreed
- Who should receive the record once it is approved
- Who is going to review it
The participant
A participant writes nothing up. Once the record is published they receive the summary, the decisions in the wording that was agreed, and their own action items with any dates attached. Someone who was invited and could not attend reads the summary to catch up. A participant who followed the call through captions in another language works from the same record as everyone else.
If a participant spots something wrong, they raise a correction against that line. The item goes back for another look instead of being edited quietly. That keeps the published version and what people believe it says as one thing. Follow-up runs from the same record: owners are reminded of the items assigned to them, and the next meeting opens on the decisions of the last one.
The reviewer
The reviewer opens the draft and reads the summary against the transcript. Names and terms the system misheard are corrected through the correction workflow. A decision recorded in wording that could be read two ways is rewritten so it reads one way. Any action item without an owner gets one or gets removed. Approval is the action that takes the record from draft to reviewed, and it carries the reviewer's name.
Distribution is confirmed at the same point. The reviewed record is published to the people who already had access to the meeting.
Why the reviewer is a separate role
Machine output varies with audio quality, with accents, and with how technical the discussion was, so what a given draft got wrong cannot be guessed in advance. Splitting the role off also helps because whoever hosted the meeting was busy talking in it, and a second reader catches the passage where the summary tidied away a caveat that mattered. The transcript is the reference both of them can point at when memories disagree. The arrangement only holds up if the reviewer has time booked for the job.
How RoomHex runs this
Meeting Capsule is the RoomHex capability that carries the three states. It produces the transcript with speaker labels, the summary, the decisions, and the action items with owners, keeps them in draft until someone approves them, and handles the follow-up afterwards. Captions and translation during the call feed the same record for teams working in more than one language.
Decisions, plans and the exact words people used are sensitive material, which is a large part of why organizations self-host in the first place. None of it leaves the deployment, and the draft state holds until a reviewer approves it.
None of it lives apart from the meeting. The record is made by the same room the conversation happened in, reached through the same domain as the call. There is no second system to grant anyone access to, and no export step between the two.
- transcripts
- action-items
- ai
- follow-up
- review