Every angle at once
Sub-agents work trade areas, ramp-up, formats, members, and operations side by side, so a boardroom-ready view arrives from one prompt.
ClaudeWren AI for Claude
Claude dashboards your CEO can trust.
One prompt sends Claude's sub-agents after every angle of your business at once. Wren AI gives them the same definitions across 20+ databases, and every number on the finished dashboard links to its SQL.
- Parallel sub-agents, one context layer
- Every number linked to its SQL
- Row- and column-level security
Store Expansion Outlook
Built by Claude on Wren AI
Transit-hub uplift
+25.2%
SQL: revenue_by_trade_area
Months to maturity
10
SQL: ramp_up_by_month
Network growth YTD
+3.5%
SQL: network_by_month
Format to open next
Compact
SQL: format_economics
Revenue per store, by trade areaSQL
Transit hub125
CBD108
Residential97
Suburban91
Parallel sub-agents
One context layer
SQL per number
From the blog walkthrough on a sample retail dataset. Trade-area values indexed to the network average of 100 and shown for illustration.
Trusted by data teams worldwide, with 17,841 GitHub stars
Parallel sub-agents
One prompt. A team of analysts, working at once.
Claude splits your question into angles and sends a sub-agent after each one. Every sub-agent reads the same Wren AI context layer, pulls its slice of live data, and reports back. Claude merges the numbers, checks them against each other, and builds the dashboard.
You ask
Claude plans the analysis
Build an executive dashboard for the CEO and CIO on our store expansion plan. Clear, visual, boardroom-ready.
Fans out to sub-agents
- Sub-agent 01in parallel
Trade areas
Where does a store earn the most?
pos_monthly_salesstore_masterTransit hubs earn 25.2% more per store
- Sub-agent 02in parallel
Ramp-up
How fast does a new store mature?
store_monthly_kpinew_store_rampupNew stores reach maturity in month 10
- Sub-agent 03in parallel
Network growth
Is the base business still growing?
pos_monthly_salesThe network is up 3.5% year to date
- Sub-agent 04in parallel
Formats and cities
Which format and city should open next?
store_masterstore_monthly_kpiCompact, high-yield formats lead
- Sub-agent 05in parallel
Members
Who shops where, and how often?
member_rfmMember segments mapped to each trade area
- Sub-agent 06in parallel
Operations
Can staffing and supply keep up?
human_resourcesinventoryStaffing and inventory checked per store
Claude merges and verifies
- Puts every angle on one rolling 12-month window
- Cross-checks the headline numbers before using them
- Keeps each figure's SQL as a live source
Store Expansion Outlook
Every number linked to SQLOpen new doors where traffic is densest, in compact, high-yield formats. Transit-hub stores earn 25.2% more per month, new stores mature in month 10, and the network is growing 3.5% year to date.
Numbers that add up
Every sub-agent reads the same metric definitions and declared joins, so the pieces agree when Claude puts them together.
Across all your databases
One MCP connection reaches 20+ sources. A sub-agent on sales and another on HR can each query the system where that data lives.
Angles shown follow the store-expansion walkthrough on a sample retail dataset.
Meet Wren AI
Why Claude needs a context layer.
A 44-second introduction to Wren AI: what it is, where it sits, and what it gives Claude when Claude works with your data.
- 01Wren AI connects Claude to 20+ databases through one MCP connection.
- 02Your context layer defines what each column means, how tables join, and the business rules, so Claude stops guessing.
- 03Every assistant your team uses works from the same definitions, and every number traces back to its SQL.
Why it matters
Generated is easy. Governed is the bar.
A dashboard built by an AI model is only as trustworthy as the path the data took to get there. Pointed at a raw schema, Claude guesses what columns mean and picks its own joins. Connected through Wren AI, it reads your definitions first.
| Dimension | Claude on a raw database | Claude on Wren AI |
|---|---|---|
| What columns mean | Inferred from names | Read from the context layer |
| Joins | Chosen by the model | Declared once, reused by every caller |
| Sources | One connection per database | One MCP connection, 20+ sources |
| Analysis | One thread, one query at a time | Parallel sub-agents that share one set of definitions |
| Access | Whatever the service account can see | Row- and column-level security per user |
| Every number | A figure in a chat | Linked to the SQL that produced it |
When someone in the meeting asks where a number comes from, the answer is one click away, not buried in a chat transcript.
What Claude reads first
The context a column name never tells you.
Definitions live in MDL, Wren AI's context model, as git-versioned files in your repository. Claude reads them before it writes a single query.
1. Meaning
Columns arrive with their meaning
In the walkthrough, floor_area_ping is store floor area in ping, about 3.3 square meters each. Without that, revenue per square meter comes out wrong by a factor of three.
github.com/acme/wren-contextmodels/store_master/metadata.yml
main1name: store_master2columns:3 - name: floor_area_ping4 description: Floor area in ping (1 ping is about 3.3 m2)Claude reads this before it writes SQL, so revenue per square meter comes out right.2. Joins
Joins are declared, not guessed
Store Monthly KPI joins to Human Resources on store and month, not on store alone. The relationship is declared once, so every caller gets the same join.
github.com/acme/wren-context/relationships.ymlstore_kpi_hr
store_monthly_kpi
- store_id
- year_month
- revenue
- foot_traffic
human_resources
- store_id
- year_month
- headcount
- labor_hours
Join onstore_idyear_monthDeclared once. Every sub-agent joins on store and month, never on store alone.
3. Access
Each viewer sees their own rows
Access rules apply at query time to the person asking. Two people who ask for the same dashboard see only the rows and columns each is allowed to see, and every request is logged.
Same request: Build the store expansion dashboard
North region manager
Row policy: region = North
- North$2.1M
- Centralhidden
- Southhidden
Head of expansion
All regions
- North$2.1M
- Central$3.4M
- South$1.8M
logged: caller, resolved plan and SQL for each request
One context layer
Every assistant your team uses, one definition of revenue.
Teams rarely standardize on a single assistant. When each tool connects to the database on its own, each carries its own idea of what the numbers mean. Wren AI defines them once and serves them to every surface, with the same access rules.
ChatGPT
ChatGPT app directory
Ask in the chat your team already works in, and get the SQL and a chart back.
Wren AI in ChatGPT →This pageClaude
SlackClaude connector, MCP
Turn one prompt into a dashboard where every number links to its SQL.
Slack app
Answer the question in the thread where it was asked.
Wren AI in Slack →Microsoft TeamsTeams connector
Bring governed answers into Teams chats and channels.
Wren AI in Teams →Your own agentAPI, MCP server
Give your product or internal agent the same context layer.
MCP server →FAQ
Claude dashboards on Wren AI, answered.
Enable the MCP connection on your Wren AI project, then search "Wren AI" in Claude's connector directory and authorize it with OAuth. A database credential never reaches Claude. The connector announcement covers setup in detail.
Claude breaks a request like "build an executive dashboard on our expansion plan" into angles, such as trade areas, store ramp-up, formats, and operations, and works them in parallel. Each sub-agent queries through Wren AI, so all of them use the same metric definitions, joins, and access rules. Claude then merges the results, cross-checks the headline numbers, and builds the dashboard with each figure linked to its SQL.
Every source Wren AI supports, 20+ in all, including PostgreSQL, SQL Server, Oracle, MySQL, Snowflake, BigQuery, Databricks, Redshift, ClickHouse, Athena, Trino, Starburst. One MCP connection reaches all of them, and the Wren AI engine plans each query against the source where the data already lives. There is no ETL step and no migration.
Yes. Every query runs through Wren AI, and every number on the dashboard links to the SQL that produced it. Each request is also logged with who asked and the SQL that ran.
Only if they are allowed to. Row- and column-level security apply at query time to the person asking, so each viewer sees only the rows and columns they are permitted to see. These controls and audit logs are part of Wren AI's paid plans.
Your context layer: what each column means, how tables join, and the business rules your team defined. These live in MDL as git-versioned files in your own repository, so a definition changed in a pull request is what Claude uses on its next question.
No. The same context layer serves ChatGPT, Slack, Microsoft Teams, and any MCP client through the Wren AI MCP server. Wren AI core is open source, so you can run the whole path locally first.
Try it
Put Claude on governed data.
Add Wren AI from Claude's connector directory, or see it on your own data with our team.