Bastion Data
Product · Parley

Ask AI in plain language. Get a chart, a grid, or a reviewed change.

Parley is an AI chat layer over the databases you already run. One conversation drives reporting, visualization, tabular editing, and bulk data management, and the model never decides what is safe to write.

See the four modes
Availability
Private · by request
Modes
Report · Table · Data Manager · Schema
Databases
PostgreSQL · MySQL · SQL Server
Your rows
Never sent to the model
The problem

The questions arrive faster than the queries.

Business intelligence tools want to copy your data into their store and have you maintain a modeling layer. Notebooks answer one analyst’s question, once. Neither lets anyone safely fix the row that is simply wrong, so corrections happen in a spreadsheet, or in psql at six in the evening.

01

Ask the one person who writes SQL

Every question about the data becomes a ticket in someone else’s queue.

02

Wait for the extract

A CSV arrives, already stale, cut off from the database it came from.

03

Fix it by hand and hope

A cleaning UPDATE runs against live data with no way to see what it will touch.

One conversation, four modes

01

Report

Ask a data question and get a visualization that answers it. The AI writes the SQL and the chart configuration: sixteen chart types, from bar and line through sankey, treemap, heatmap, and geographic maps. Pin the ones that prove useful into folders the team or public sharing.

02

Table

A spreadsheet-like grid over any query, with editors derived from your schema: enums become dropdowns, foreign keys become pickers that show a label and store an id, and a null renders as a distinct rather than an empty cell.

03

Data Manager

Describe a bulk change, like "archive every survey still pending from before 2020", and read a row-by-row diff of exactly what it would do, cascades and your own triggers included, before deciding whether to apply it.

04

Schema

Browse tables, views, and columns, and manage the data dictionary directly in the database as comments, where every other tool can read it too. Slice and dice access to your database via in-tool schemas and expose only what you want the AI to understand.

Why it matters

The model proposes. Your schema disposes.

A language model is a capable writer and a poor authority. Parley lets it do the generative part, turning a question into a candidate query, while a deterministic engine derives every safety-critical answer from your schema’s own keys, relationships, and grants: what is editable, how to write it back, what a change would touch. A bad guess produces a wrong-but-harmless result or a read-only grid. It cannot produce silent corruption.

Schema and intent sent to the modelyes
Row data sent to the modelnever
Preview transactionrolled back
Write-back not derivableread-only
Sovereignty

It attaches to your database. It never absorbs it.

Parley connects out to a database you own and keeps its own state (conversations, pinned views, audit trail) on its side of the line. Nothing tool-owned is written into your data: no extra columns, no bookkeeping rows, no permanent triggers. The change preview runs inside a transaction that is rolled back, so it leaves your schema byte for byte as it found it.

Your grants are the boundary. What the connection role is permitted to do is exactly what Parley can do, and the database enforces that, not our code.

Parley chat pins audit
Your database tables views grants

Lift the software off and a complete, rational dataset remains. You lose your pins, not your data.

What Parley does

Ask & see

Question to chart

Plain language in, SQL and a rendered visualization out.

Sixteen chart types

Bar through sankey, treemap, heatmap, and maps.

Pinned and shared

Save the shapes that prove useful into shared folders.

Edit & correct

Schema-driven editors

Dropdowns, pickers, and null handling derived from your types.

A visible change trail

Every edit stays marked, and anything marked can be restored.

Diff before apply

Bulk changes are reviewed as rows, not trusted as SQL.

Govern & document

Grants decide capability

Read, write, and preview follow the role you connect with.

Audit outside your tables

Who changed what, recorded without touching your data.

Documentation that stays

The data dictionary lives in the database, readable by any tool.

Who Parley is for

Your team waits on you for every number

You can write the query in two minutes, but you are the queue, and the questions never stop arriving.

Data leadsProgram managers

Someone capable is editing production directly

Not recklessness: raw SQL is simply the fastest road under a deadline. The paved road has to be faster, and safer.

Database administratorsSelf-taught power users

The data is not yours to hand over

You hold it under a data-sharing agreement, so portability and non-lock-in are contract terms, not preferences.

Research data custodiansCompliance owners

Why not just use an existing tool?

ParleyBI platformsNotebooksRaw SQL
Runs on the database you already ownNativeCopies to its store
Natural-language questionsSometimes
Corrects data, not just reads it
Row-level diff before a write
Safety derived from the schemaYour judgement
Modeling layer to maintainNoneRequired
Leaves nothing installed in your databaseVaries

Managed hosting

No infrastructure to run. Parley connects out to your database over a credential you issue, and that you can revoke in a single statement.

On-premises / self-hosted

For residency, compliance, or governance requirements. Runs entirely inside your infrastructure, against a model provider key you control.

Private availability

Curious what it would make of your schema?

Tell us what your data looks like and what you keep having to ask it. We’ll show you the round trip.

or email hello@bastion.dev