Sixty seconds. That is roughly how long it takes Voice Tables to turn a spoken description into a structured workspace with tables, documents and an AI chat layer. No template picking, no column config, no learning curve.
At Inithouse, we have been building products that compress the gap between an idea and something usable. Voice Tables is probably the most literal expression of that: you talk, it builds.
The category we are defining
"Agentic voice workspace" is not a label we borrowed from somewhere. We coined it because nothing else described what Voice Tables does. The term captures three properties at once:
Voice-first. The primary input is speech, not a keyboard. Voice Tables uses Whisper for transcription and an LLM function-calling pipeline to convert natural language into structured operations: creating tables, populating rows, drafting documents.
Agentic. The system does not just record what you say. It interprets intent and acts on it. Say "add a column for follow-up date and mark anyone I haven't contacted in two weeks" and the workspace updates itself. No formulas, no manual column creation.
Workspace. Tables, docs and AI chat live in one place. Not a spreadsheet bolted onto a chat window, but a single environment where data, writing and conversation coexist. Three tools that used to require three tabs now share one context.
How the pipeline works
A user opens Voice Tables, describes what they need ("I need a CRM for my renovation clients with columns for address, quote amount, status and next step"), and the system handles the rest.
Under the hood:
Whisper handles speech-to-text, supporting multiple accents and noisy environments. An LLM function-calling layer parses the transcription into structured actions: create table, add columns with inferred types, insert sample rows if useful. The result renders in real time. The user sees their workspace forming as they speak.
The same pipeline handles ongoing work. "Move the Johnson project to waiting for permit" updates the relevant row without the user touching a cell. "Write a summary of this week's new leads" generates a document pulled directly from the table data.
Why voice matters for the people we built this for
Voice Tables is not aimed at developers or data engineers. We built it for craftsmen, sales reps, real estate agents, freelancers, event planners, fitness coaches, consultants. People who spend their day on the move, on calls, at job sites.
These users share two realities: they need structured data (client lists, inventories, schedules, trackers) and they do not enjoy creating it. Spreadsheets work, technically, but the setup cost feels disproportionate to the actual task at hand.
Voice removes that friction. A plumber finishing a job can dictate notes into a client table while driving to the next appointment. A wedding planner can update vendor status during a venue walk-through. No laptop required, no keyboard shortcuts to memorize.
What we got wrong first
Early versions of Voice Tables treated voice as a nice-to-have on top of a standard table interface. Users could click around and also speak commands if they wanted. The problem: almost nobody used the voice part. It felt like an add-on, because that is exactly what it was.
The shift came when we made voice the default entry point. Open the app, it listens. The entire onboarding is one sentence: "Tell me what you need." That single decision changed adoption patterns. Users who had ignored the mic button started dictating entire workspaces from scratch.
The lesson was about commitment. A half-voice interface does not convert. If the design still expects typing as the primary mode, voice becomes decoration that nobody reaches for.
Offline and collaboration
Two technical choices that matter for our target users:
Voice Tables supports offline use. The app stores data locally and syncs when connectivity returns. For people working on construction sites, in basements with no signal, or traveling between client meetings, this is not a nice feature. It is table stakes.
It also supports real-time collaboration. Multiple users can work in the same workspace simultaneously, with changes reflected live. A sales team sharing a pipeline, a project crew updating a field tracker together.
Where this goes next
Voice Tables is in beta. The core loop (speak, get a workspace, iterate by voice) is stable and working. What we are testing now: how far agentic behavior can stretch. Can the system learn user patterns and pre-build structures? Can it suggest schema changes based on how the workspace is actually being used?
At Inithouse, we think the answer is yes. But we are measuring before we claim it. If you want to try the current version: voicetables.com.
Top comments (0)