Start the agent from a role you trust
An owner creates the agent from an existing role, usually analyst, and the agent gets its own copy, named after it. Editing the base role later does not change the agent, so it never gains access by accident.
Gateway
Gateway decides who can query what in your warehouse. Each person signs in with their own role, and each AI agent gets a role, a token and a daily call limit of its own. Roles limit the schemas, columns and rows a caller can read, and every query lands in a log under the caller’s name.
Each agent gets a copy of a role you already trust, named after it. Owner-level roles are never offered as a starting point.
The agent connects with its own token and a daily call cap. It cannot sign in to Rig, and a leaked token is replaced in one step.
The role sets the schemas, tables, columns and rows the agent can read, and PII policies mask sensitive fields.
Every call lands in the audit log with the agent’s name and tool, including the calls Gateway refused.
People connect Claude, Claude Code, Cursor, Codex or ChatGPT to Rig over MCP and sign in as themselves, so each tool sees exactly what that person may see. Agents that a team shares, such as a Notion custom agent or a Slack assistant, connect with an agent token instead.
Before a query runs, the tool looks up the certified metric in Rig Map and Gateway applies the caller’s role. The warehouse password stays with Rig.
What’s our net revenue retention for enterprise accounts last quarter?
Used Rig integration
search_metricsnet_revenue_retentionget_metric_detailscustomer × quarterexecute_sql1 row returnedNRR for enterprise in 2025 Q4 was 112%.
Resolves to your certified net_revenue_retention metric, filtered to segment = 'enterprise' for period = '2025Q4'. Rows scoped per RBAC.
An owner creates the agent from an existing role, usually analyst, and the agent gets its own copy, named after it. Editing the base role later does not change the agent, so it never gains access by accident.
Rig shows the MCP URL and the token once, for pasting into Claude, Notion, Viktor or any MCP client. The daily call cap is 1,000 by default and resets at midnight UTC. A leaked token is replaced in one step, and the old one stops working at once.
The Access Matrix sets which datasets a role reaches. Permissions set table, column and row rules, and PII policies mask sensitive fields. Owners can run SQL as a role and see exactly what comes back before granting it.
The audit log shows every call with the agent’s name and the tool it ran in. Refused calls show as errors, and switching an agent off refuses its next call on every server.
A disputes agent that reads payments data and nothing else
A sales assistant in Slack that every rep can use without seeing the admin’s view
Claude and Cursor on certified metrics across the company, under one set of rules
Customer story / Cleo
One Cleo Growth Specialist’s agents made 8,023 requests to company data in 30 days, on the certified metrics Cleo’s data team reviews.
Read the story“Every team has a Rig agent now. Ops, support, fraud, marketing. We stopped being the bottleneck.”
Certify the metrics once in Rig Map, then connect Claude or Cursor to Rig’s MCP server. A person signs in with OAuth and gets their own role. An agent gets its own role and token. Either way, the tool looks up the certified metric before it writes SQL, sees only the schemas, columns and rows the role allows, and every call lands in the audit log.
It is a layer between AI tools and the warehouse that identifies each caller, applies its role’s schema, column and row rules, limits how often it can call, and logs every query. Rig’s Gateway does this for people as well as agents.
Create the agent in Rig from a role you trust, narrow what it can reach, set its daily call cap, then paste the MCP URL and token into the agent’s tool. It cannot sign in to Rig, it never appears in your user list, and every call is logged against it by name.
Any MCP client can connect, including Claude, Claude Code, Cursor, Codex and ChatGPT. Notion custom agents, Viktor and Claude Tag connect with an agent token.
No. Those tables need to know which person is asking, and an agent has no personal identity, so Rig refuses the query rather than guessing. Give the agent a role with its own row rules instead.
It works with Snowflake, BigQuery, Databricks, Redshift, Postgres and the other warehouses Rig connects to, or with a warehouse Rig hosts for you. On Databricks, Unity Catalog grants and column masks carry through.
Rig Map holds the definitions an agent reads before it writes SQL, and Gateway decides what that agent may query. Both run on the warehouse you already have.
Talk to Rig
Tell us where your team spends time, loses money or needs better information.
In a 30-minute call, we’ll discuss the data and systems work that could help.