← All integrations

Data

Give your agents full access to your databases.

Connect Postgres, MySQL, or ClickHouse and agents introspect the schema, write real SQL, and reason over the results in their own sandbox. Reach private databases through an SSH tunnel without opening anything to the internet.

5,000 free credits to start.

Agent reach

Postgres, MySQL, and ClickHouse over SQL

The full surface, searched at run time.

Connect with

  • Connection string

Add an SSH tunnel with a password or private key for databases on a private network.

Event triggers

Not yet

Manual, schedule, and signed webhook are available.

What the agent can do

What agents do with Databases

Agents introspect the schema, write SQL, and work over the results in their own sandbox. Credentials stay outside the sandbox.

  • Connect Postgres, MySQL, and ClickHouse with credentials stored envelope-encrypted.
  • Introspect schemas - tables, columns, types, keys, and indexes - before writing a query.
  • Write and run SQL against your data, then analyze results in the sandbox with code.
  • Reach private databases through an SSH tunnel, so nothing needs public exposure.
  • Grant read-only credentials so a reporting agent can never write.
  • Queries run read-only with a 60 second timeout and a 100 row default limit.

Examples

Tasks you can hand to a database agent

Ask your database a question in English

The agent reads the schema, writes the SQL, runs it, and explains the answer - including the caveats in the data.

Which accounts grew usage more than 50% last month, and what do they have in common?

  1. postgres·introspect schema for usage tables
  2. postgres·query month-over-month usage by account
  3. code·cluster the growing accounts by attributes

23 accounts grew over 50%

  • 18 connected a second integration last month
  • Median team size 12, up from 7
  • Query and assumptions included in the run

Publish a metrics report on a schedule

Every morning the agent runs the queries, computes the derived metrics, and posts a report your team reads.

Weekdays at 7:00am

Each morning, post yesterday's metrics with anything unusual called out.

  1. clickhouse·query events for the last 24h
  2. postgres·join account metadata
  3. code·compute trends and flag anomalies
  4. slack·post report to #metrics

Metrics · Mar 6

  • Runs 41.2k · up 6% day over day
  • Signups 128 · in line with the trend
  • Error rate up 0.4pp, concentrated in one region

Run data quality checks every night

The agent checks the invariants you care about, and when one breaks it says which rows, since when, and what changed upstream.

Nightly at 3:00am

Nightly, run our data quality checks and alert on failures.

  1. postgres·run 14 invariant checks
  2. postgres·isolate failing rows and first occurrence
  3. resend·email the on-call analyst

2 checks failed

  • `orders.currency` null on 41 rows since 02:10
  • `accounts.plan` has 3 values outside the enum
  • 12 checks passed

Move data where it needs to be

Query results become CRM fields, spreadsheet rows, or an email. The agent handles the shape change in code.

Nightly at 1:00am

Nightly, sync product usage into the CRM as account health fields.

  1. postgres·compute 30-day usage per account
  2. code·map usage into health tiers
  3. attio·update 480 account records

480 accounts synced

  • 31 moved into `at risk`
  • 18 flagged as expansion candidates

Triggers

How a database workflow starts

Databases does not emit event triggers yet. Workflows that use it start by hand, on a schedule, or from a signed webhook.

Manual

Run it from the app with an optional JSON payload. Every workflow supports this.

Schedule

A cron expression or a plain sentence such as every weekday at 9am. Times are UTC.

Signed webhook

A public URL per workflow. Requests are signed with HMAC-SHA256 and expire after five minutes.

Works with

Your databases work with the rest of your stack

One agent, one run, several services. It reads from one tool and writes to the next in the same sandbox.

FAQ

Questions, answered

Put an agent on your databases.

Connect it once and describe the task in chat. Then reuse the prompt as a workflow on a schedule or a webhook.