platform / ai-agent

Ask your data anything. Get answers you can trust.

The agent answers questions about your data in plain language, shows its working and draws charts. It works from your semantic layer, so its numbers match your dashboards.

request-demo read-the-docs

agent — ask

show weekly revenue by channel for the last 90 days, as stacked bars

✔ table: revenue_daily · metric: revenue
✔ period: last 90 days (assumed)
✔ processed 1.2 GB · 13 rows
chart: stacked bar · download PNG

now add a trendline and ignore the partial week

trusted by data teams at

camino/financial

✓ Enterprise-ready. Already trusted by multiple Fortune 100 companies.

see it

Chat with your data and get charts you can verify

Ask in plain language, follow up to change the chart, and expand any step to check the query, the SQL and the data processed.

agent — chat

Which traffic sources brought the most users in the last 30 days?

2 steps · processed 0.1 GB

Users by first-user source, last 30 days

top 5 source / medium

Google organic brought the most users, followed closely by direct. Together they make up most of the traffic. Period assumed: last 30 days. Metric: users (from the semantic layer).

make it weekly share as 100% stacked bars

Share of weekly users by channel

google organic · direct · other

google organicdirectother

how-it-answers.md

// steps · expand any step to verify

step 1  query  users by first_user_source
         metric from semantic layer · period: last 30 days (assumed)
sql     select source_medium, count(distinct user_id)
         from marts.users_daily where date >= current_date - 30
rows    5   processed 0.1 GB (limit 10 GB)

charts.md

Advanced visualisations

line

bar

stacked

donut

scatter

trendlinesmoving averagesrunning totalsreference linessecond axisvalue labelsPNG download

// illustrative

what you get

AI agent, in detail

plain-language.md

Plain-language answers

Ask from the Agent page or the ask box on Home, then follow up to change the breakdown or the chart.

accurate.md

Built on your semantic layer

It uses your dbt models, metrics and dimensions, plus your knowledge documents, so numbers match your dashboards.

shows-work.md

Shows its working

Every answer has steps with the query, SQL, rows returned and data processed. It states assumptions such as the time period.

charts.md

Advanced visualisations

Line, area, bar, stacked, pie, donut and scatter, with trendlines, moving averages, running totals, reference lines and a second axis. Download as PNG.

safe.md

Safe and cost-controlled

Only SELECT queries run. Oversized queries are blocked, date-partitioned tables are always date-limited and a monthly spend cap applies.

your-llm.md

Your choice of model

Use your own account with OpenAI, Anthropic or Google. Keys are stored encrypted, and you can change the model any time.

how it works

From setup to answers

01 /

Set up

An admin adds an API key for OpenAI, Anthropic or Google, picks a model and sets a monthly spend cap.

02 /

Ask

Editors and admins ask in plain language, from the Agent page or the Home ask box.

03 /

Check

Expand the steps to see the query, the SQL and the data processed.

04 /

Share a chat with the project so teammates can read it.

anywhere

Prefer to work somewhere else?

The same governed data is available outside the platform through our MCP server, so you can use it in Claude, Slack, your editor or your own agents, with the same benefits.

faq

Frequently asked questions

How does the agent get its numbers?

From your dbt models: the tables, metrics and dimensions exposed in Explore. That is why its numbers match your dashboards. It also reads your knowledge documents for definitions.

Can I trust the answers?

Each answer shows its steps: the query, the SQL, the rows returned and the data processed. It states assumptions such as the time period. It can still be wrong, so check the steps for important results.

Which AI models does it use?

Your choice of OpenAI, Anthropic or Google, using your own account. You can change the model at any time.

What happens to my data?

Questions and the query results go to the AI provider your admin selected, under your own account with them. Reflective Data's own AI accounts are not used.

Can it change or delete data?

No. Only SELECT queries run, and it cannot see tables outside your models.

How is cost controlled?

Queries over the project size limit are not run, date-partitioned tables are always date-limited, and an admin sets a monthly spend cap.

Who can use it?

Editors and admins, once an admin has set it up. Viewers do not see it.