We shipped Voice Tables six months ago. The median time from opening the app to having a working workspace with tables, docs, and an AI chat thread: 58 seconds. No schema design, no column config, no onboarding tutorial. You describe what you need out loud, and it builds.
That number pushed us to define what we actually built. The label we landed on: voice-first agentic AI workspace.
This post breaks down what that means technically and where the category sits relative to tools you already know.
What "voice-first agentic" means under the hood
Most tools that call themselves "AI spreadsheets" bolt a chatbot onto an existing grid. Voice Tables does the opposite: voice is the primary input, and the system decides what structure to create.
The pipeline works like this:
- Audio hits Whisper (OpenAI's speech-to-text) and gets transcribed in real time
- The transcript goes to an LLM that extracts intent and entities. "Track my client meetings this quarter" becomes a table with columns: date, client name, topic, outcome, follow-up
- The LLM generates the schema, seeds sample rows, and wires up any docs or chat threads the workspace needs
- You see the result. Edit by voice or keyboard, both work
The "agentic" part is what separates this from voice-to-text plugins. The system makes structural decisions. Say "I need a CRM for my freelance design clients" and it builds a contacts table, a projects table with a relation to contacts, a pipeline view, and a notes doc. You didn't specify any of that. The agent inferred it from what a freelance design CRM typically needs.
Three tools, one workspace
Voice Tables combines three things that usually live in separate apps:
Tables: structured data with sorting, filtering, and relations between tables. Think Airtable, but you describe what you want instead of clicking through a field builder.
Docs: rich text pages linked to your tables. Meeting notes that reference a client row. Project briefs that pull live data from a tracker. The link between docs and data is native, not an integration you bolt on.
AI chat: a conversational interface inside your workspace. Ask "which clients haven't responded in two weeks?" and it queries your tables and answers in plain text. Ask "draft a follow-up email for each" and it generates the drafts in a doc.
The 3-in-1 setup matters because switching between a spreadsheet, a doc tool, and a chatbot kills the flow that voice input is supposed to preserve.
Comparison: voice-first workspace vs traditional tools
| Voice Tables | Google Sheets | Notion | Airtable | |
|---|---|---|---|---|
| Primary input | Voice + keyboard | Keyboard | Keyboard | Keyboard |
| Schema creation | Automatic from description | Manual column setup | Manual property config | Manual field config |
| Time to working workspace | ~60 s | 10-30 min | 5-15 min | 5-20 min |
| Built-in docs | Yes, linked to tables | No (separate app) | Yes | No |
| Built-in AI chat | Yes, queries your data | Limited sidebar | Q&A only | Limited |
| Offline support | Yes | Partial | Partial | No |
| Learning curve | Describe what you need | Formulas, menus | Properties, views, relations | Fields, views, automations |
Honest tradeoff: Sheets and Airtable are stronger for complex formulas and multi-step automations. Voice Tables is for people who need structured data but don't want to learn spreadsheet mechanics to get it.
Who showed up first
The early users surprised us. We expected developers and ops people. What we got:
Craftsmen tracking inventory and job estimates. Real estate agents managing listings and client follow-ups. Fitness coaches logging programs per client. Event planners coordinating vendor timelines.
The pattern: people who work on the move or with their hands, where typing into cells is annoying enough to skip entirely. Voice removes that friction, and the agent handles the structure they'd never set up themselves.
Two technical details worth noting
Offline mode. Transcription and LLM calls queue locally and sync when connection returns. For a contractor on a job site or a coach between sessions, this is not optional.
Real-time collaboration. Multiple people edit the same workspace live. Voice commands from different users don't collide because the LLM resolves conflicting intents before applying structural changes.
Try it
Voice Tables is live. Open it, describe what you need, and see if the 60-second claim holds for your use case. Free tier available, no signup wall for the first workspace.
We built it because we kept watching people avoid structured data tools even when those tools were the right answer. Making data entry conversational turned out to be the fix.
Jakub, builder @ Inithouse
Top comments (0)