Setting up a CRM from scratch usually takes somewhere between an afternoon and a whole sprint. You pick a tool, drag columns around, wire up automations, realize the schema was wrong, and start over. We wanted to see what happens when you skip all that and just describe what you need out loud.
Voice Tables is an agentic AI workspace we built at Inithouse. You talk to it. It builds the tables, fields, and relations for you. Here's what the CRM walkthrough actually looked like.
The prompt was one sentence
We opened Voice Tables, hit the mic button, and said:
"I need a CRM to track leads, companies, and deals. Each deal should belong to a company and have a stage pipeline."
That was it. No drag-and-drop. No template gallery. No config wizard with seven tabs.
What Voice Tables built in about 40 seconds
Three linked tables appeared:
Leads: name, email, phone, source channel, status (new / contacted / qualified / lost), linked company, notes, created date.
Companies: company name, industry, size range, website, linked leads (auto-populated), linked deals.
Deals: deal name, value, stage (discovery / proposal / negotiation / closed-won / closed-lost), expected close date, linked company, assigned owner.
The relations between tables were already wired. Clicking a company showed its leads and deals inline. The stage field on Deals came with a Kanban view by default.
We didn't pick any of those field names from a dropdown. The workspace inferred them from the voice input and from the fact that CRMs tend to follow predictable data patterns. The result was closer to what we would have built manually than we expected. The "source channel" field on Leads, for example, was something we'd have added in round two.
What we changed afterward
Two things. We added a "last contacted" date field to Leads (voice command: "add a date field called last contacted to leads") and renamed the "size range" column on Companies to "employee count", again just by saying it.
The point isn't that the first output was perfect. It's that getting from zero to a working schema took under a minute, and every adjustment was the same interface: describe the change, see it happen.
Why this matters for small teams
Most CRMs are either too simple (a shared spreadsheet) or too complex (multi-week onboarding, admin certifications, dedicated ops person). The gap in between is where voice-first data modeling gets interesting.
A freelancer tracking 30 clients doesn't need deal stages. A three-person agency does. With Voice Tables, both describe what they need in their own words and get a workspace shaped to fit. No spreadsheet formulas, no admin panel, no YouTube tutorial on how to set up custom fields.
We've seen similar patterns across other use cases at Inithouse: fitness coaches tracking client programs, event planners managing vendor contacts, consultants logging project hours. The schema is different every time; the interaction pattern is the same: say what you need, watch it appear, adjust by talking.
The technical bit
Voice Tables runs speech-to-intent through a local pipeline. Your voice input gets transcribed, parsed into structured actions (create table, add field, set relation type), and executed against the database layer. The workspace supports offline use and syncs when you reconnect. Real-time collaboration means two people can talk to the same workspace without overwriting each other's changes.
For the CRM use case specifically, the pipeline handled the three-table-with-relations structure in one pass. That's worth noting because multi-table schemas with foreign keys are where most no-code builders start requiring manual configuration.
What we'd do differently
If we rebuilt this walkthrough, we'd start with the company table first and let the workspace infer that leads and deals are child entities. The current approach (saying everything in one sentence) works, but starting from the "biggest" entity and building outward tends to produce cleaner schemas when the data model gets more complex (say, 6-8 linked tables).
That's an observation from about three months of internal use. Not a rule, just a pattern that keeps showing up.
Voice Tables is built by Inithouse. If you've tried building a CRM by talking to it, or used voice input for any structured data problem, we'd like to hear how it went.
Top comments (0)