Surface 01
Canvas
A grid of pinned, versioned views your whole team shares. A pipeline board and a metric row is the whole CRM most small teams need.
Entity graph · event log · claims with provenance
Wavetable replaces the CRM, the CMS, and the ops-tool sprawl with one graph of your customers, revenue, content, and traffic. AI writes the configuration. The runtime is deterministic. Every answer carries a receipt.
Ten minutes, one founder, zero training.
The problem
The CRM has one version of a customer. The email has another. The billing system has a third. None of them can tell you where a fact came from, whether it's still true, or what changed it. So every report is an argument and every automation is a leap of faith. Wavetable starts from a different physics: facts are claims with provenance, and everything you see can show its work.
The first ten minutes
This is the actual first-run path, not an aspiration.
Sign up. Connect Gmail and Stripe. Paste a 2KB site script into your website. That is the whole setup.
People, companies, revenue, and pageviews arrive as events, become claims, and get deduplicated into one graph. Each record carries a written narrative in which every sentence is cited.
Ask the command bar a real question. The answer renders as a view with the exact query behind it. One click pins it to your team's Canvas.
Describe recurring work in plain language. Before it goes live, Wavetable replays it against your real history and shows you the counterfactual, with the actual drafts it would have sent.
Replay report · quiet-trial check-in
This would have fired 14 times last quarter. Here are the 3 drafts.
Tomorrow there are two drafts waiting, one merge proposal with evidence, and one schema suggestion. Approve, hold, or reject. Every decision is an event with your name on it.
Your workspace exposes a policy-scoped MCP endpoint. Point Claude at it and ask it anything about your business, within the policies you approved, with every call logged.
One app, five surfaces
Surface 01
A grid of pinned, versioned views your whole team shares. A pipeline board and a metric row is the whole CRM most small teams need.
Surface 02
Ask anything in plain language. The answer arrives as a view with a receipt: the exact query, the rows, the plan hash. Keep it? Promote it to the Canvas.
Surface 03
Every person, company, and deal carries a written narrative where every sentence is cited, backed by a claims table with full provenance.
Surface 04
The only place work waits for you: drafts to approve, merges with evidence, artifact diffs with replay reports. Built to clear in five minutes a day.
Surface 05
Your schema, views, workflows, and policies with git-like history. Diff any two versions. Rollback is a button.
Trust, structurally
Provenance
Every fact carries its source events, its asserter, a confidence, and a status. The tables you query are rebuildable caches of accepted claims. "Where did this number come from?" always has an answer.
Citations
The narrative on every record maps each sentence to the claims behind it. A sentence the generator cannot cite does not ship. Prose without provenance is not allowed in the system of record.
Replay
Every automation is tested against your real history in shadow mode first. You see how often it would have fired and what it would have sent, before it can touch anything.
Governance
No agent takes external action without a capability minted by a policy you approved. The limits are structural, enforced by the runtime, never a prompt asking a model to behave.
The architecture bet
The agent is a compiler, not the runtime. It authors your schema, views, workflows, and policies as versioned artifacts you approve. What runs every day is compiled, inspectable, and reproducible.
Design plane · AI, slow, human-gated
Describe what you want in plain language. The Compiler drafts the artifact, runs it against your history, and hands you a diff with a report. Expensive model calls happen here, where a human checks the output.
Run plane · deterministic, fast, cheap
Approved artifacts execute as compiled programs: same input, same output, every time, with budgets and blast-radius limits enforced by the runtime. Your Tuesday does not depend on a model's mood.
Packs
A pack bundles schema, views, workflows, and policies into one approval. Everything it ships becomes ordinary artifacts you own: forkable, diffable, reversible.
Pack · agency-gtm
Companies, deals, a pipeline board, an open-pipeline metric, a stale-deal follow-up, and a gated welcome email for new form fills. A working pipeline in one session.
Pack · content
Posts with editorial states, an editorial board, scheduled publishing, and traffic attribution. Published posts serve over a public content API, straight from the substrate.
Agent-native
Every workspace exposes a policy-scoped MCP endpoint, so any agent you authorize (Claude, a partner's agent, your own scripts) can query your graph, read cited memory, and propose work. Agents come and go; the substrate is where they all return. Lock-in comes from your data and artifacts living here, not from walls.
Security
One database per tenant, no cross-tenant query path anywhere in the codebase. Erasure works by crypto-shredding: shred the keys, redact the claims, rebuild the projections. Every approval, policy change, and agent call is an audited event.
Connect Gmail, Stripe, and your site. Ask a question, get a receipt, and approve your first automation today.
Get started freeNo credit card. Connect your data, get receipts in minutes.