Xata

Let AI agents work your Xata serverless Postgres data. Agents insert, update, and bulk-load records, run typed queries, full-text search across tables and branches, aggregations, summaries, and raw SQL.

13 actions API key available

Xata website ↗ Platform Documentation ↗ Capability data verified 2026-07-27

The nightly sync from the order system finishes exporting

Bulk Insert Records loads the incoming rows in a single request, with Xata generating ids

Aggregate Records reconciles row counts and totals against the source system's figures

Query Table pulls the specific records behind any mismatch for inspection

The agent drafts the corrective SQL and rehearses it with Run SQL on a working branch to measure affected rows

The data owner reviews the statement and its affected-row count before it runs against main

The reconciliation summary posts to [Slack](/integrations/slack)

What This Integration Enables

Xata's bet is that Postgres should behave like a product: serverless, branchable, searchable, with typed data APIs alongside raw SQL. That combination is what makes it a database agents can work responsibly. A FlowRunner agent gets structured record operations for the routine writes, filter queries and full-text search for the reads, aggregations and summaries for the reporting, and parameterized SQL for everything the structured API cannot express. Branches are the safety architecture: what would be a scary statement on production becomes a rehearsal on a copy, and the [human-in-the-loop](/concepts/human-in-the-loop) approval happens with evidence in hand. - Insert, upsert, update, and bulk-load records with schema-matched JSON - Query with structured filters and run fuzzy full-text search across one table or the whole branch - Compute counts, sums, averages, and grouped summaries for reporting without an export - Execute parameterized Postgres-compatible SQL for joins and advanced work

Without FlowRunner

Every question is a ticket Operational data exists, but answering from it means finding an engineer with console access

Imports are one-off scripts Each load is a bespoke loader with no validation and no record of what it did

Fixes run raw against production Corrective SQL gets pasted into a console late in the day, WHERE clause willing

With FlowRunner

Agents answer from the data Queries, aggregations, and grouped summaries run on schedule and land where people read

Loads reconcile themselves Every bulk insert is followed by an aggregate check against the source totals

Writes are rehearsed Statements prove out on a branch with row counts shown, and a person approves the run on main

Use Case Scenarios

The metrics digest nobody compiles

Each morning the agent runs Summarize Records on the orders table, grouped by status and channel with sums and averages per group, plus Aggregate Records for the day's headline counts. The result posts to [Slack](/integrations/slack) as a digest and appends to [Google Sheets](/integrations/google-sheets) for the running history. When a number looks off, the follow-up is immediate: Query Table with the same filter pulls the underlying records into the thread, so the conversation happens over rows, not recollections.

External systems synced without column loss

Rows flow in from [Google Sheets](/integrations/google-sheets) and the order platform. For records keyed on an external id, the agent deliberately avoids Create or Replace Record for routine syncs, because a replace clears every column the payload omits. Instead it uses Update Record with upsert enabled: existing rows get partial updates that preserve untouched columns, and missing rows get created. Full replaces are reserved for loads that genuinely own the whole record, and Bulk Insert Records handles first-time imports at volume.

One search box over every table

Support gets a customer email and needs everything: orders, tickets, shipments, notes. The agent runs Search Branch with the customer's email and name, and gets back relevance-ranked records tagged with their source table, across the entire branch in one call. The assembled picture posts into the support thread within seconds, and for compliance and data-subject requests the matched records archive to [S3](/integrations/s3) as a dated, auditable bundle. Nobody writes a five-table join to answer a routine lookup ever again.

Human-in-Loop Highlight

Run SQL executes whatever statement it is given against the branch, and on main that includes the UPDATE with the missing WHERE clause that rewrites a column for every row in the table. Delete Record is explicit about being permanent, and Create or Replace Record silently clears any column the payload leaves out, which is data loss wearing an upsert's clothing. The FlowRunner posture uses Xata's own architecture as the gate: destructive statements rehearse on a working branch first, the agent reports exactly how many rows the statement touched there, and the run against main waits for the data owner to read the statement and the count and say so. A database that makes branching cheap has removed every excuse for unrehearsed writes to production.

Agent processes routinely

Detects exception requiring judgment

Clear match Continues automatically

Ambiguous Routes to human via preferred channel

Human decides

Agent resumes with decision

Agent Capabilities

13 actions

Records

Insert Record Inserts a record with Xata generating the id, from a JSON object matching the table schema. The default write for new rows.
Create Record Creates a record with a specific id and fails if that id exists. The guarantee to reach for when overwriting would be the bug.
Create or Replace Record Upserts by id, fully replacing an existing record. Columns not in the payload are cleared, so it belongs to loads that own the whole record.
Get Record Retrieves a record by id, with column selection including dot notation into linked tables like author.name.
Update Record Partially updates only the provided columns, optionally creating the record if it is missing. The safe default for syncs.
Delete Record Permanently deletes a record by id, optionally returning selected columns of what was removed. Behind confirmation, always.
Bulk Insert Records Inserts many records in one request with generated ids. The import primitive for seeds and nightly loads.

Query

Query Table Queries with structured filter objects, sorting, column selection, and cursor or offset pagination. The read behind reporting and record inspection.

Search

Search Table Fuzzy full-text search over one table's searchable columns, relevance ranked, with prefix matching and column boosting.
Search Branch Full-text search across every table in the branch at once, results tagged with their source table. The one-call customer lookup.

Analytics

Aggregate Records Computes counts, sums, averages, min and max, unique counts, and date histograms, optionally filtered. The reconciliation and dashboard primitive.
Summarize Records Groups records SQL-GROUP-BY style with per-group summary functions, filterable and sortable on the output. One row per group, ready for a digest.

SQL

Run SQL Executes raw Postgres-compatible SQL with $1, $2 parameter placeholders against the branch. Joins, advanced aggregations, and the corrective statements that only run on main after a person approves.

Frequently Asked Questions

What can FlowRunner do with Xata?

FlowRunner agents can run Insert Record, Create Record, and Create or Replace Record in Xata, plus 10 more actions.

Does connecting Xata to FlowRunner require OAuth?

No. Xata connects to FlowRunner with an API key, no OAuth flow required.

Can Xata trigger a FlowRunner workflow automatically?

Xata doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.

Start building with Xata

$100 in credits. No card required. Connect in minutes.

Start Building Free Talk to us