Ask in plain English. No SQL required.
Conversational analytics for PostgreSQL where every follow-up shows its SQL.
Google's data agents are the deeper fit when the data already sits in Cloud SQL or AlloyDB. Chion works on the Postgres you already run (RDS, Supabase, Neon, or self-hosted) and answers a follow-up by editing the prior query instead of starting over. The SELECT that produced each chart stays on the page, read-only and capped at 1,000 rows.
7-day trial. Connect a read-only PostgreSQL role.
What is conversational analytics?
Query data by chatting in plain English. No SQL, no dashboards.
You ask a question in plain English, the system generates and runs a code-validated SQL query, and returns a chart plus a written explanation. Each follow-up question keeps the context of the previous turn, so you can refine results by saying "break that down by quarter" or "exclude churned accounts." Unlike Google's data agents (scoped to Cloud SQL and AlloyDB) or spreadsheet tools, the code-validated SELECT that produced the chart is shown beneath it.
How conversational analytics works
Three steps: connect, ask, read the chart and the SELECT.
- Connect your database. Paste a read-only PostgreSQL connection string. Credentials are stored in an encrypted vault and never leave Chion's backend.
- Ask a question. Type it the way you'd ask a colleague: "which customers churned last quarter and what did they spend?"
- Read the chart and the SELECT. Chion returns an interactive D3 chart, the exact SQL it ran, and a written summary. Ask a follow-up to refine.
Every answer shows its SQL and its meaning.
Grounded narratives built from actual query output, not generic summaries.
Most analytics tools hand you a chart and leave you to figure it out. We generate a structured narrative for every result: a grounded analysis built from the actual query output, not a generic summary.
Every narrative follows a mandatory arc: a headline that names all series with their direction and explicit date ranges, a key insight that identifies cross-series patterns like shared peaks or divergences, and per-series statistical facts: non-competing observations that each add new information.
Narrative depth scales with the data.
Simple charts get short answers. Complex data gets deeper analysis.
Not every chart needs the same level of narrative depth. A single-series bar chart with 5 rows doesn't need three paragraphs. We compute a data density tier deterministically in code (not by the LLM) and use it to gate which narrative fields get populated.
Thin data
1 series, fewer than 10 rowsSummary only. Headline and key insight are set to null.
Moderate data
2–3 seriesQA (question alignment) + summary. Key insight populated when cross-series patterns exist.
Rich data
4+ series or high row countAll three fields: headline, key insight, and per-series facts. Full statistical analysis.
Context carries across every turn.
Follow-up questions refine the prior query. No starting over.
Each conversation tracks entities and filters across turns. When you say "break that down by quarter," the system knows you're refining the prior analysis; it doesn't start from scratch. It uses keyword delta tracking to detect what changed between your current question and the previous one.
Filter changes are classified into four operations: add (new filter that didn't exist before), replace (same dimension, different value), subtract (remove a filter), and preserve (carry forward unchanged). The SQL gets rebuilt surgically: only the parts that changed get regenerated. Everything else carries over from the previous turn's code-validated query.
This is why follow-up questions are fast. We're not re-profiling your schema or re-resolving entities on every turn. We're applying a delta to a known-good query. See how Chion compiles a code-validated SQL pipeline step by step.
Take your saved queries anywhere: export to Claude Code and Codex
Chion also offers one-shot text-to-SQL generation to skip the conversational layer and paste code-validated SQL into your own tools, plus an AI SQL analyst for ad-hoc questions.
Need a portable agent file? Chion exports your saved queries as CHION.md plus a SKILL.md per role, which Claude Code and Codex read from their skills folders.
Part of the Chion AI SQL workforce, built on the same code-validated SQL agent architecture.
How Chion compares to BI copilots
How Chion differs from BI dashboards and cloud-locked data agents
| Feature | Chion | Tableau AI | Power BI Copilot | Looker |
|---|---|---|---|---|
| Capacity or license prerequisite for the AI assistant | A Chion seat and a read-only PostgreSQL role | A Tableau Cloud site on Tableau+, with AI in Tableau turned on in site settings and a Creator or Explorer site role | Paid Fabric F2+ or Power BI Premium P1+ capacity, alongside the per-user license | Not publicly documented |
| Generated code shown to the user | The executed read-only SELECT, shown with the chart | A performance recording shows the text of each Executing Query event: SQL when connected directly to the data source, XML for a published data source | Fabric data agent generates SQL, DAX, or KQL and exposes the intermediate code | The SQL tab in the Data panel shows what Looker sends to the database, with Open in SQL Runner and Explain in SQL Runner links |
| Sources one AI agent can be configured against | Your PostgreSQL databases, connected read-only | Not publicly documented | Fabric data agent supports up to five configured sources | A Conversational Analytics Explore data agent chats with as many as five Explores at once |
| Published result caps | 1,000 rows and 12,000 cells per result | Not publicly documented | Not publicly documented | Explore queries have a maximum limit of 5,000 rows, up to 50,000 where an admin raises it |
| Portable skill artifact | Saved SQL compiled to CHION.md or SKILL.md for Claude Code and Codex | Not publicly documented | Microsoft publishes Power BI Agentic skills | Not publicly documented |
| Billing unit | Per seat, from $29 per month; Enterprise per team, custom-quoted | Not publicly documented | Per-user license, plus Fabric or Premium capacity billed separately for Copilot | Not publicly documented |
Every competitor cell traces to that vendor’s own documentation, checked 2026-08-31. Power BI: learn.microsoft.com/power-bi/create-reports/copilot-introduction, microsoft.com/power-platform/products/power-bi/pricing, learn.microsoft.com/fabric/data-science/how-to-create-data-agent, learn.microsoft.com/fabric/data-science/data-agent-runtime, and learn.microsoft.com/power-bi/developer/agentic/power-bi-agentic-overview. Tableau: help.tableau.com/current/online/en-us/web_author_einstein.htm, help.tableau.com/current/server/en-us/perf_record_interpret_server.htm, help.tableau.com/current/pro/desktop/en-us/perf_extracts.htm, and help.tableau.com/current/online/en-us/pulse_create_metrics.htm. Looker: cloud.google.com/looker/docs/how-looker-generates-sql, cloud.google.com/looker/docs/best-practices/row-limits-in-looker, and cloud.google.com/looker/docs/conversational-analytics-looker-data. "Not publicly documented" means the dimension is absent from the sources listed for that product on that date, not that the capability is missing. Chion figures are this site’s published plan prices and its current result caps. Correct a cell by opening a PR against src/data/comparisons.ts.
See our vault-encryption and read-only SQL security model → · Full Chion vs BI dashboards comparison →
Frequently asked questions
9 answers about conversational analytics.
What is conversational analytics?+
Conversational analytics lets you query data by typing plain-English questions instead of writing SQL or navigating dashboards. The system generates a code-validated SELECT, runs it against your database, and returns an interactive chart with a written explanation.
How is conversational analytics different from traditional BI dashboards?+
Dashboards show pre-built views that someone else configured. Conversational analytics lets you ask any question on the fly: no ticket, no analyst, no wait. Each answer includes the SQL so you can verify it.
Do I need to know SQL to use Chion?+
No. You ask questions in plain English. Chion generates the SQL, checks it against a typed contract in code, and runs it. The SQL is visible if you want to inspect it, but you never need to write it.
What databases does Chion support?+
Chion connects to PostgreSQL via a read-only connection string. Your credentials are stored in an AES-256-GCM encrypted vault and never leave Chion's backend.
Is my database data safe? Can Chion write to my tables?+
Chion enforces read-only SELECT queries with a hard row limit. It cannot INSERT, UPDATE, DELETE, or ALTER your data. Every query passes a read-only check before execution.
How does Chion check the SQL before running it?+
Each generated query passes through a multi-phase validation pipeline: schema alignment, contract enforcement, and a read-only safety check. Invalid queries are rejected before they reach your database.
Can I see the SQL Chion generated?+
Yes. Every chart includes the exact SQL that produced it. Click the query to view, copy, or run it independently in your own database client.
What happens when I ask a follow-up question?+
Chion tracks entities and filters across turns. When you say 'break that down by quarter,' it refines the prior query surgically: only the changed parts are regenerated. Everything else carries over from the previous turn's code-validated query.
How is Chion different from Google's conversational analytics for Cloud SQL and AlloyDB?+
Chion works on any PostgreSQL (RDS, Supabase, Neon, or self-hosted), shows the exact SQL under every chart, enforces read-only with a 1,000-row cap, and exports every answer as a portable agent file. Teams fully committed to GCP with data already in AlloyDB get native IAM integration from Google's agents; that remains the deeper choice inside Google's cloud.
Point Chion at your Postgres and ask.
Point it at your database and ask something. See what comes back.