Voice · Structured Capture

Murmur

Speak a messy thought. Get structured records — in the tools you already use.

In short

Murmur is the missing input layer: talk the way you actually think, and get structured records in the tools you already use.

▸ExpandCollapse

Every CRM, task manager, and note system has the same weakness: it's only as good as what people bother to type into it. How often do you diligently log the five people you met this morning, with who they were and what you promised them?

The cost isn't just a stale database. It's the follow-ups that never happen — the opportunities that quietly expire. And if you work inside an organization, it's everything you learned that the organization never gets to act on.

Murmur is the missing input layer. You talk — messy, unstructured, the way you actually think — and an AI turns it into structured records: tasks with due dates, people matched to contacts you already have, notes linked to the right project. It doesn't replace your system; it feeds it.

Live demo — press Speak and use your own scenario Say something like the sample below: a messy after-meeting brain dump with people, promises, dates, and context. When you start speaking, the sample output is replaced by follow-ups extracted from your words. Nothing is saved.

Mic recording works in modern mobile and desktop browsers; . Nothing is saved.

SampleThis is example input and output. Press Speak or type your own; the boxes below will be replaced.

Sample transcript — For example, you might say… "Just got back from the chamber networking breakfast — met Priya Raman who runs ops at Corewood Logistics, she wants an intro to a fractional CFO by next week. Also talked to Dave Okafor, startup guy, send him that article about agentic coding. And remind me Thursday to follow up with the table sponsor about the fall event."

Example output · 2 Contacts

Priya RamanOps · Corewood Logistics · met 7/3
Dave OkaforFounder · new contact

Example output · 3 Tasks

Intro Priya → fractional CFOdue Jul 10
Send Dave the agentic-coding articleno date
Follow up: table sponsor, fall eventdue Thu Jul 9

Example output · 1 Note

Chamber networking breakfastlinked: Priya, Dave · Jul 3

Speak

Stream of consciousness in. No forms, no fields, no friction.

Structure

An LLM extracts tasks, people, dates, and relationships — and shows its work live.

Match

"Priya" fuzzy-matches the Priya you already know. Ambiguity gets flagged, not guessed.

Review

Everything saves instantly; anything worth a second look waits in a review queue.

Under the hood

Local-first writes, invisible sync. Records save immediately, then a background engine syncs them to the destination system — Notion in my own setup, but the pattern could fit a CRM, task manager, or notes app.
Deterministic guardrails around a nondeterministic model. Entity positions are re-anchored by string search (LLMs can't count characters); hallucinated field values are coerced or rejected; every parse has a retry and a fallback model.
Streaming, phase by phase. You watch extraction happen — entities appear as they're found, not after a spinner.

◆ Invited guests see more: the sync-engine design, the review-queue UX in depth, and where this is headed next. Unlock →