Meetings

What a meeting is

A meeting in Elliptic is a conversation turned into a searchable, source-anchored record. Bring in a transcript and Elliptic writes the summary, links every line back to the exact moment it came from, turns action items into tasks, answers follow-up questions, and lets you share the result with people outside the workspace. A meeting can carry a transcript, your own notes, or both, and everything stays connected to the rest of your org so a task can point back to the meeting it was decided in.

Every AI step here runs on your organization's own model key (your OpenAI or Anthropic key, stored once at the org level). Summaries, asking one meeting, asking across meetings, and guest questions on shared links all execute on that BYOK key, so the cost lands on your provider bill and nothing leaves your control.

Open Meetings from the left sidebar of your workspace. You land on the meeting list, newest first, with three actions in the top-right: Ask across meetings, New meeting, and Import meeting.

Who can see a meeting

A meeting is either filed under a project or left org-wide. That single choice controls visibility:

• Org-wide meetings (not filed under any project) are visible to everyone in the organization.
• Project meetings are visible to members of that project, plus org admins and owners.
• Org admins and owners see every meeting, filed or not.

To attach a meeting to a project you must be a member of that project. Org admins and owners can attach to any project. This keeps project-specific conversations visible only to the people on that project, while general meetings stay open to the whole org.

The auto-filing suggestion

When a meeting is not filed yet, a suggestion banner appears at the top of its page. Elliptic scores the meeting against your active projects and proposes the best match with a confidence band (a 0-to-1 score under the hood). Accept it to file in one click, choose another active project from the picker, or dismiss the banner to keep the meeting org-wide. If there is not enough signal to guess, the banner skips the suggestion and simply asks you to pick a project. Filing matters because it sets who can see the meeting, and it is required before you can turn action items into tasks, since tasks live inside projects.

Getting a meeting into Elliptic

There are two ways to create a meeting: import an existing transcript, or start a blank meeting and write notes by hand.

Import a meeting (the main path)

Click Import meeting to open the importer. It has three tabs, so you can use whatever you already have on hand:

1. 1
   Open the importer

   Click Import meeting on the Meetings page, or from the empty state if you have no meetings yet.
2. 2
   Pick your source

   Choose Paste transcript, Folio JSON, or Upload file depending on what you have.
3. 3
   Add a title

   For pasted text or an uploaded text transcript, give the meeting a title such as "Weekly sync". A Folio JSON paste carries its own title, and an uploaded .json export takes its title from the file.
4. 4
   Import

   Click Import meeting. Elliptic creates the meeting and its transcript segments together, then opens it.

Start a blank meeting

Click New meeting, give it a title ("Standup", "Customer call", "1:1"), and press Create meeting. This opens an empty meeting with no transcript required. Use it for a live or ad-hoc session and type your own notes as you go. You can summarize and chat once there is something to work with, either a transcript or your notes.

Attendees

A meeting tracks two kinds of attendees: members of your organization (linked by their account) and external attendees recorded as plain names. Folio imports carry their attendee list across as external names. All member attendees must belong to the organization, and the source badge on the meeting tells you where it came from, Folio or Manual.

The meeting page

Opening a meeting shows its title, source badge, date, and duration. From here you work in two views: the Document tab and the Transcript tab. A Split view button puts the document and transcript side by side on a wide screen, and an Ask panel stays docked on the right.

The Document tab

The Document tab is the readable version of the meeting: the AI summary up top, and your own notes at the bottom. It is where you generate, refine, and adopt a summary, and where you turn action items into tasks.

The Transcript tab

The Transcript tab is the verbatim record: every segment with its speaker, timestamp, and text, in order. This is the ground truth that AI summary lines and chat citations link into, so you can always check a claim against what was actually said. It loads in pages for longer meetings with a Show more control. When you click a source link from a summary line or a chat citation, the transcript scrolls to that exact segment and highlights it.

Auto chapters

Longer transcripts get a chapters rail of 2 to 8 auto-generated topic jump points, each labelled from the start of its section and timestamped. Click one to jump down the transcript, and the active chapter highlights as you scroll. Chapters are derived from the transcript segments, so short meetings simply do not show them.

Split view

On a wide screen, click Split view to show the document and transcript at the same time, so you can read the summary on one side and verify against the raw transcript on the other without switching tabs. Click it again to exit.

Deleting a meeting

The meeting's creator or an org admin can delete a meeting. Deleting removes the meeting and its transcript together. Like every other meeting action, the deletion is recorded in the org activity log.

AI summaries

Click Summarize (or pick a template first, then Summarize) and Elliptic reads the transcript on your org's AI key and produces a structured summary. Each line is grounded: the model returns its summary as discrete lines, each tagged with a section and the transcript segment ids that support it. The result is grouped into labelled sections with counts:

• Decisions — what was decided.
• Action items — what someone needs to do, each carrying an "Action item" badge.
• Open questions — what was raised but not resolved.
• Key points — the rest of the highlights.

Provenance and source anchoring

Every AI line is shown in dimmed text so you can tell at a glance what the model wrote versus what you wrote, since your own notes render in full contrast. Each line carries a small source marker. Hover it to see the speaker, the timestamp, and the exact transcript quote behind that line. Click it to jump straight to that moment in the Transcript tab. The header tells you how many lines are linked to the transcript (for example "7 of 9 lines linked to transcript"), and a line with no clear source is honestly flagged for you to verify manually instead of carrying a fake citation.

Summary templates

Templates shape how the AI organizes a summary. Pick a template from the dropdown next to the Summarize button before you generate, and the summary is structured around that template's sections, in order. There are built-in templates for the common meeting shapes:

Org admins can create custom templates with their own named sections and an optional extra instruction (a prompt scaffold) passed to the AI on top of the section list. A custom template might define sections like "Risks, Mitigations, Owners" plus an instruction such as "Flag anything that needs legal review." Once saved, it appears in the same template dropdown for everyone in the org, under a name that must be unique. Admins can rename, re-section, or delete custom templates. The built-in ones are always available and cannot be removed.

Re-enhance and summary history

• Re-enhance — regenerate the AI summary from the transcript, optionally under a different template. Any summary lines you adopted and edited as your own are preserved, so re-running never wipes human work.
• Summary history — each run is saved with the model and provider it ran on, newest first, so you keep a record of how a summary evolved and on which model.

Your notes

The bottom block of the Document tab is always yours to write. Notes are source material, not AI output, and a blank meeting can be built entirely from notes.

• Your notes — click Edit to add or change notes in a rich editor. They render in full contrast and are clearly marked as yours.
• Edit as your notes — copy the AI summary down into your notes so you can edit it. Once it lives in your notes it reads as your own words, marked "Edited by you", and is no longer treated as raw AI output.

Turning action items into tasks

This is how a meeting stops being a document and starts driving work. Action items in the summary each get a Task button, and the section header has a Create all tasks button. Filing a single action item creates one task; Create all tasks creates one task per action item. Each task is created in the project's backlog with no priority, and it remembers which meeting it came from, so the work stays linked to its source.

1. 1
   Make sure the meeting has a project

   Tasks live in projects. If the meeting is filed in a project, tasks go there automatically. If it is not, a "File tasks in" picker appears so you can choose a destination project for this session.
2. 2
   Create one task, or all of them

   Hover an action item and click Task to file just that one, or click Create all tasks in the Action items header to file every action item at once.
3. 3
   Track it like any other task

   The new tasks appear in that project's backlog, linked back to this meeting so you always know where the work came from.

Ask about one meeting

Every meeting has an Ask panel docked to the right (focus it instantly with ⌘J / Ctrl+J). Ask anything about what was discussed, decided, or assigned, and the AI answers strictly from that one transcript, on your org's key. Answers are ephemeral: they are not saved into the meeting document, and they are designed to stay grounded.

• Preset prompts — one-click starters like "What did I miss?", "Summarize the last 5 minutes", "List decisions made", and "Suggest questions for me to ask".
• Citation chips — when an answer is grounded, numbered source chips appear under it. Hover one to see the quote and timestamp, click it to jump to that segment in the transcript.
• Honesty signal — if the answer reads as low confidence (for example "that was not discussed"), it is flagged with a warning to verify directly. Grounded answers carry a quieter "check the transcript to verify" note.
• Action lines to tasks — if a line in the answer matches an action item in the summary, a Task button appears on that line so you can file it on the spot.

Recipes: saved transcript prompts

Type / in the Ask box to open the recipes menu. Recipes are saved instructions you run against the meeting, grounded in the transcript just like a normal question, so they will not invent facts the meeting does not contain. You can save your own, and any recipe you save becomes available on every meeting in the org.

To save a recipe, type / followed by the instruction you want to keep, then choose "Save … as a recipe". It then appears in the / menu on every meeting in the org under that instruction's name. Running a recipe is recorded in the activity log.

Ask across all meetings

The single-meeting Ask panel answers about one conversation. Ask across meetings (top-right of the Meetings list) answers about the whole archive. Use it for questions like "What did we decide about the API redesign this quarter?" Elliptic scans your most recent visible meetings (up to 50), ranks them by how well they match your question, pulls the best-matching passages from the strongest few, and answers from those excerpts only.

• Citations — every answer lists the meetings and moments it drew from, as links you can open.
• Coverage — it tells you how many meetings it consulted out of how many it scanned, so you know how broad the answer is.
• Scope filters — narrow the search to a single project, or to a date range (From / To), before you ask.
• Save as page — turn any answer into a Note in one click, titled from your question, so a good cross-meeting synthesis becomes a permanent, shareable page.
• Respects visibility — it only ever searches meetings you are allowed to see.

Sharing and Slack

There are two ways to get a meeting out to people who are not looking at it inside Elliptic: a public share link, and a push to Slack.

Public share links with guest Ask

From the meeting page, the creator or an org admin can mint a public link that needs no login. Guests always get the summary, action items, and decisions. The full transcript is a separate, opt-in tier you toggle on with Include transcript. Guests also get an Ask about this meeting box that runs on your org's key, scoped only to what you shared, so they can never reach beyond this one meeting. Revoking the link cuts access instantly without deleting the record.

Post a summary to Slack

Once Slack is connected for your org, click Send to Slack on the meeting, pick a channel, and Elliptic posts a tidy message: the title, the latest summary, the action items as a bulleted list, and, when an active share exists, a link teammates can click to ask the AI about the meeting. If Slack is not connected yet, the dialog points you to your org settings to connect it first.

Agents work meetings too

Elliptic is Jira for your agents, so everything on this page is available to your AI members over the company-brain MCP, on the same org key. An agent can import a Folio recording (import_folio_meeting), summarize on a template (summarize_meeting), and read what a person reads: chapters (list_meeting_chapters), summaries with their model and provider (list_meeting_summaries), and transcript segments (list_meeting_segments).

From there an agent can ask one meeting (ask_meeting) or the whole archive with coverage and citations (meetings_chat), run a saved recipe (run_meeting_recipe), file action items as linked tasks (create_task), define templates and recipes (create_meeting_template, create_meeting_recipe), get a filing suggestion (suggest_meeting_project), mint a share (create_meeting_share), and post a recap to a channel (post_meeting_to_slack). The visibility rules, the BYOK key, the anti-hallucination guarantee, and the activity log all apply identically whether a person or an agent does the work.

Everything is logged

Meeting actions are recorded in your org's activity log: created, imported, updated, summarized, recipe runs, shared, share revoked, posted to Slack, and deleted. So you always have a trail of how a meeting was used and who used it, person or agent.