What it does
- Paste a meeting transcript.
- Review what the agent pulled out for each ticket it heard about: updates, action items and who was involved. Edit or remove anything.
- Confirm each ticket by swiping to accept or reject it.
- Post. Approved updates land in Jira as comments, new subtasks and “is blocked by” links between issues.
The demo above runs on a sandbox Jira project with made-up tickets.
How it works
Two choices shaped it:
- A locally run model. Meeting transcripts are internal, so the LLM runs on the laptop through Ollama instead of a hosted API, and the transcript never leaves the machine.
- A person approves every write. The agent proposes and a human confirms, ticket by ticket, before anything touches Jira. A wrong blocker link or a stray subtask costs more trust than the time it saves.