Pilot program open — limited spots for this quarter.Apply
The platform

One platform. The whole loop.

Six AI specialists work as one system. Three surface the decisions your business acts on. Three are the data foundation they run on. All grounded in your own systems, with a human approving every consequential move.

Start a pilot How it works
The loop you talk toAskInvestigateDashboards·The foundation underneathArchitectPipelinesForecast
The loop you talk to

Observe. Investigate. Recommend.

The three specialists a decision-maker talks to directly — they watch what matters, dig into why, and hand back the move. No SQL, no waiting.

Ask

Self-serve answers

Type any business question in plain English and get a sourced answer back in seconds — grounded in your data, no dashboards, no SQL.

Ask
Sales feel off this month — why?
Ad-driven orders are down 12% — Meta CPMs jumped, ROAS fell below target.
Answer appears here, grounded in your data

Investigate

Root-cause analysis

Give it a hard question and watch it plan hypotheses, test them against your data, and hand back a ranked, finding-backed report.

Investigate
Top 3 ways to increase profitability?
Proposing 3 hypotheses — approve & run.
Key finding: Laptops carry 15.1% margin, 1.7pp below blended.
Ranked, finding-backed strategy ready →
The investigation streams here as it runs

Dashboards

Living dashboards

Describe what you want to understand and it builds a living dashboard — widgets appear as you talk, and it refreshes itself.

Dashboard Builder
Orders aren't growing — which channel is the problem?
Building: orders by channel, trend, and returns.
Published — it refreshes itself daily.
Your dashboard takes shape as you talk
The foundation underneath

What the loop runs on.

The three specialists that connect your systems, model them into one governed picture, and forecast ahead — so every decision above is fresh, grounded, and traceable.

Architect

Semantic model

Connect your ERP, POS and more; it profiles every table and models them into one clean, governed semantic layer — your source of truth.

Data Architect
Connect NetSuite & Shopify.
Found 42 tables. Classifying facts & dimensions…
Modelling into one canonical layer.
Published — one governed source of truth.
Sources become one clean, governed model

Pipelines

Data pipelines

Ask for fresh data and it builds the pipeline — source, transform, test, and schedule — no hand-written SQL or cron jobs.

Pipeline Builder
Keep daily sales fresh across all channels.
Reading sources → cleaning → unifying.
Added tests — row counts & nulls look good.
Scheduled. Runs every night at 02:00.
1
Source
2
Transform
3
Validate
4
Schedule
Validation✓ 128,402 rows · 0 nulls
Schedulenightly · 02:00 · Temporal

Forecast

Forecasting & ML

It picks a model, trains on your own data, validates against a quality gate, and serves forecasts — demand, churn, risk.

Model Trainer
Forecast demand for the next 8 weeks.
Selected LightGBM on 327K sales lines — training…
Validated: WAPE 6.2%, beats baseline. Gate passed.
Serving forecasts — refreshed weekly.
Train models on your own data
How it fits together

One platform, five layers.

Raw data enters through connectors, becomes one clean picture, and the agents turn it into decisions — with governance and per-industry packs wrapped around every layer.

Governance & Trust
HITL approval · tenant isolation · grounding
Business
The answer, ready to act on
AskInvestigateDashboards
Agents
Does the work of a data team
6 specialist agentsreason · plan · execute
Agent tools
~180 purpose-built tools
querycatalogpipelinesMLdashboardsingest
Core infrastructure
The engine room
ProcessingML training + inferenceTemporal orchestration
Data layer
One clean, grounded picture
Canonical + semantic modelLakehouse
Feed connectors
40 sources — ERP · POS · supplier · marketplace
Vertical Pack Engine
reusable templates & playbooks per industry
Governance & trust

Every agent runs inside the guardrails.

Human-in-the-loop

Agents propose; a human approves anything that writes or acts.

Read-only by default

Analysis agents read to answer — they never change your data.

Grounded, not guessing

Every answer is tied to your real data, with its source shown.

Per-tenant isolation

Your data and models are yours alone — never shared or trained on.

See the whole loop on your data.

Start a pilot — bring one problem, and watch datameld connect, investigate, recommend, and measure it in two weeks.

Start a pilot How it works