Guide · Capture and extraction

Turn meeting notes into tickets: a practical guide

To turn meeting notes into tickets, capture the source, review the extracted drafts with their evidence quotes, approve the ones that hold up, then push them to a tracker. Relay keeps the original sentence attached to each draft, so every ticket can answer where it came from.

By the Relay maintainers · Last updated 19 November 2026 · Source on GitHub

What kind of notes work?

Relay accepts plain text and Markdown notes, DOCX and PDF files, VTT and SRT transcripts, transcript JSON and email in .eml form. Notetaker exports can be sent through the capture API. Audio is supported too, but it needs a transcription endpoint or a transcript supplied alongside it. For a first run, pnpm demo --file notes.txt walks the whole pipeline with zero credentials and a fake tracker. Hand-written notes with clear decisions and owners give the cleanest drafts.

How do you turn meeting notes into tickets?

  1. Capture. Post the notes in a watched Slack or Discord channel, mention the bot in a thread, use /relay log, or send a file to the API.
  2. Extract. Relay drafts a title, type, priority, owner and due date, each linked to a quoted line.
  3. Review. A person checks the quote, edits the fields and decides whether the draft is real work.
  4. Push. Approval queues the draft in an idempotent outbox that creates one tracker issue.

Why check the evidence quote?

A draft without its source is a guess dressed as a task. Relay shows the exact line that produced each draft, with its speaker and timestamp where the source has them. That lets a reviewer confirm the commitment is real and resolve ambiguous language before it becomes an issue. If the quote does not support the draft, reject it with a reason; the rejection is stored as learning feedback.

In an illustrative note, the line “we need a retry with backoff on the billing webhook” produces a draft titled “Add retry to the billing webhook,” shown with that same quote.

Illustrative demo data.

How is an owner chosen?

Relay suggests up to three candidates with reasons and a confidence score. The documented signals are expertise from past work, recent context from the conversation, availability and a fairness share. You can ask why a name appeared and override it. Every manual change is kept as feedback. The score is there to start a decision, not to make it for you.

What happens before a ticket is pushed?

Nothing reaches a tracker until a person approves it. Reviewers can approve, edit, reject, delegate, merge or split a draft, and low-risk drafts can be batched under policy. Approved tickets go through a database-backed outbox with replay protection, so a retry or restart produces one issue, not two. The tracker link is stored and posted back into the thread that produced it.

What are the common mistakes?

  1. Skipping review because the draft looks plausible.
  2. Losing the source quote when the note is summarised away.
  3. Guessing an owner with no explanation a reviewer can check.
  4. Treating live model output as deterministic; it varies across runs.

Extraction on messy multilingual fixtures scores about 0.78 precision and 0.75 recall, so plan to edit some drafts.

Next step

Run the credential-free demo with one of your own notes, then self-host Relay from the repository. Hosted cloud plans are planned, not live; self-hosting is the working path today.


Self-host Relay Read the workflow guide