Gentility

Docs

Connect company data.

Connect a database and your AI client, then publish a saved query for a colleague. Saved queries currently require a direct PostgreSQL connection. For general database access on a private network, you can use a connector instead.

A saved query gives an AI client a reviewed way to ask a database a question.
Illustration · A saved query gives an AI client a reviewed way to ask a database a question.
1

Create your organization

Sign in at gentility.ai with Google or your identity provider, and create an organization. Everything that follows — machines, databases, published queries, the audit log — belongs to it.

2

Install a connector if you need one

Skip this step for a direct database connection, including the saved-query walkthrough below. For a database on a private network, install the connector package (github.com/gentility-ai/agent). Install it on a server that can reach the databases you want to expose. Linux (Debian/Ubuntu) and macOS are supported.

These instructions install the database agent, which uses gentility auth. The 6net connector uses gentility login; it is a different program with the same command name. Check which program your shell finds with command -v gentility before authorizing it. Do not replace one program with the other on a machine that relies on it.

Quick install

curl -sSL https://install.gentility.ai | bash
gentility auth

Debian / Ubuntu (apt)

sudo mkdir -p /etc/apt/keyrings
curl -s https://packages.gentility.ai/gentility-packages.gpg | sudo tee /etc/apt/keyrings/gentility-packages.asc > /dev/null
echo "deb [signed-by=/etc/apt/keyrings/gentility-packages.asc] https://packages.gentility.ai/debian/ stable main" | sudo tee /etc/apt/sources.list.d/gentility.list
sudo apt update && sudo apt install gentility-agent
gentility auth

macOS (Homebrew)

brew tap gentility-ai/agent
brew install gentility-agent
gentility auth

gentility auth opens your browser to authorize the connector against your organization — no tokens to paste. Authorization does not start the agent. Complete the start step below, then check that the machine is connected in your console before configuring its database.

Start the database agent

On macOS, run the agent in a terminal and leave it running while you use the connection:

gentility run

On Debian or Ubuntu, start the installed service and enable it for future boots:

sudo systemctl start gentility
sudo systemctl enable gentility
sudo systemctl status gentility

Check both the running process or service and the connection status in Gentility. If the machine does not connect, inspect the agent's output or service logs before proceeding. To stop it, press Control-C in the macOS terminal or run sudo systemctl stop gentility on Linux.

3

Connect a database

In the console, open Databases and paste a connection string. Gentility parses it and pre-fills the connection form. Two ways to route it:

  • — Direct — for databases reachable from Gentility over the network (with an IP allowlist on your side).
  • — Through your connector — for databases on private networks. Traffic routes through the connector you installed in step 2; the database never needs to be reachable from the internet.

Use a database account with only the permissions this connection needs. Keep write protection enabled for read-only work. For a direct connection, configure TLS with certificate verification; encryption on the database connection is not automatic.

Click Test connection, then save.

4

Connect your AI

Gentility speaks MCP at https://mcp.gentility.ai/. Each person connects their own client and authorizes it with OAuth. The tools they can use depend on their organization's permissions and grants.

Claude Code

claude mcp add --scope project --transport http gentility https://mcp.gentility.ai/

This adds Gentility to the current project. Open Claude Code in that project, run /mcp, select Gentility, and follow the browser sign-in steps. Check that the server is connected before asking it to use your data. If you already added it, use /mcp to check or renew the connection instead of adding it again.

Other MCP clients

In a client that supports remote HTTP MCP servers and OAuth, add https://mcp.gentility.ai/ as a server and complete sign-in. Client availability, subscription requirements, and administrator settings vary. Follow your client's current setup instructions.

Then just ask: “Show me my databases in Gentility.”

5

Publish and share a saved query

You need permission to manage Catalog queries and read the source database. Use a PostgreSQL database connected directly to Gentility: connector-routed databases and MySQL are not supported by saved queries yet.

  1. Create a draft. Ask your AI to use catalog_query_create with your organization, database, a stable slug, a title, a description, and read-only SQL. Declare any parameters in the same order as the SQL's $1, $2 placeholders. Set a row limit appropriate to the question.
  2. Preview and review. Use catalog_query_preview with sample values for every parameter. Inspect the SQL and returned columns and rows. Publication requires a preview of the version you are publishing; a successful preview alone does not establish that the data is suitable to share.
  3. Publish. Use catalog_query_publish. Gentility checks that the SQL is read-only, that parameters match, and that you have read access to the database.
  4. Grant access. Use catalog_query_grant with the query ID and either a member's email or an access-group slug. Publishing alone does not grant access to everyone.
  5. Run it as the recipient. Have your colleague connect their MCP client. A query with the slug revenue_by_month appears as query_revenue_by_month. They supply its declared inputs; the tool runs the published SQL.

This grant lets the recipient run the saved query without general database access. Any database access they already have remains in effect. Check both direct and group grants when reviewing permissions.

Maintain access as the work changes

Use catalog_query_runs to review executions and catalog_query_grantees to inspect access. Revoke a person's or group's grant with catalog_query_revoke. If you change SQL or parameters with catalog_query_update, preview and publish the new draft before it can run.

For connection or setup help, contact [email protected].