Query your database in plain language. See how the answer was queried.
Type a question like "which plan churns the most?" instead of writing SQL. Valv gives the coding agent on your machine structured, read-only tools for your database. Its query calls stay visible in the conversation, and every answer cites the query behind each number.
A question becomes a visible query call.
The agent discovers the schema you have allowed, then sends a structured query through Valv. Expand the tool row to inspect its input and see exactly what was counted.
Starter has the most cancellations in this result.
The result stays traceable.
Rows can appear as a table or an inline chart in the conversation. Successful queries retain the complete query and result behind their citations, so a follow-up can recover what produced the answer.
Starter has the highest count:
From connection to a grounded answer
Connect
Add PostgreSQL, MySQL or MariaDB, SQLite, ClickHouse, or experimental MongoDB. Credentials stay encrypted on your machine.
Set AI access
Every column in the active schemas is available by default. Hide sensitive tables and columns before the first prompt.
Ask
The agent discovers the allowed schema and uses Valv's structured, read-only tools to answer.
Choose the output
Keep the answer in chat, ask for a live dashboard, or set up a flow that checks the number on a schedule.
Plain-language questions you can ask
Ask the way you would ask a teammate. You don't need table names. The agent looks them up in the schema you have allowed.
"How many people signed up last week?"
Finds the users or accounts table, counts rows by created date, and states the date range it used.
"Which plan churns the most?"
Groups cancellations by plan and returns the counts, with the query cited next to the answer.
"How many accounts used exports in the last 30 days?"
Counts distinct accounts in your events table for that feature and window.
"Which customers have not logged in for a month?"
Filters on the last-seen column and lists the matching rows. Hidden columns like emails stay out.
Where plain-language queries go wrong
Text to SQL is rarely wrong about syntax. It is wrong about meaning. These are the usual causes, and what helps.
Your words mean something specific
"Active user" or "churned" has a definition at your company that the schema doesn't contain. Tell the agent once and memory keeps what each metric means for the next question.
Similar tables or columns
With both users and
accounts, the agent can pick
the wrong one. Hide tables it shouldn't use in AI access, and check
the cited query when a number looks off.
Vague time ranges
"Last week" can mean the last seven days or the previous calendar week. Say which one, or read the filter in the expanded query row.
Choose what the answer becomes
Valv keeps the default lightweight. Dashboards and flows appear only when you ask for them.
Answer in the conversation
Best for a question, a follow-up, or an inline chart. The chart holds the result rows from that moment rather than rerunning later.
Scheduled flow
Ask for a flow when a number needs watching. It checks the number on a schedule, like weekdays at 08:00, and notifies you when it crosses a threshold. Flows run on your machine while Valv is open.
Live dashboard
Ask for a dashboard when the view should stay useful. The saved HTML file reruns its structured queries whenever you open it.
The agent gets read-only database tools.
The agent can inspect allowed schema and run structured queries. Those tools do not expose inserts, updates, deletes, or arbitrary SQL. Valv has no SQL editor or table editing either, so nothing in the app writes to your database.
Access starts broad, so review it first.
Valv makes every column in the active schemas available to AI by default. Use AI access to hide customer details, payment fields, or entire tables before asking a question.
Your database password stays local. The analysis may not.
Valv runs database queries from your machine and stores credentials with operating-system encryption. The coding-agent provider still processes the context needed to answer your question.
On your machine
Database credentials, the connection, and query execution stay local to Valv.
Through your provider
Your prompt, allowed schema, and relevant query results go through your existing Claude Code, Codex, Copilot, Cursor, Grok, or OpenCode account.
Text to SQL questions
What does it mean to query a database in plain language?
You type a question the way you would ask a colleague, like "how many people signed up last week?", instead of writing SQL. An AI model reads your schema, turns the question into a query, runs it, and answers with the result. This is often called text to SQL or natural language querying.
Do I need to know SQL?
No. You ask in plain language and read the answer. Each answer cites the query behind the number, so someone who does read SQL can check how it was counted.
How does text to SQL work in Valv?
You ask a question in plain language. The coding agent discovers the schema you have allowed and calls Valv's structured, read-only query tools. The agent never runs raw SQL. It sends a structured query that Valv validates and compiles, and the answer cites the query behind each number.
Can I reuse a query later?
Yes. Every answer cites the query behind each number, so you can see what ran. To keep a view current, ask for a live HTML dashboard that reruns its queries when you open it. To watch a number, ask for a flow that checks it on a schedule and notifies you when it crosses a threshold.
Can Valv create charts and dashboards?
Yes. An inline chart shows the result rows from that moment in the conversation. If you ask for a dashboard, the agent can create an HTML dashboard that reruns its structured queries when you open it. Neither file is created unless you ask for it.
Can the agent change my database?
No. The agent only gets read-only tools. It has no insert, update, or delete tools and never runs raw SQL. Valv itself has no SQL editor and no table editing, so nothing in the app writes to your database.
Which databases can I query in plain language?
Valv supports PostgreSQL, MySQL and MariaDB, SQLite, and ClickHouse. MongoDB support is experimental.
What can the agent access by default?
By default, every column in the active schemas is available to the agent. Open AI access after connecting and hide any tables or columns the agent should not use before you send your first prompt.
What data reaches my AI provider?
Database credentials stay on your machine and queries run through Valv locally. The prompt, allowed schema details, and query results needed for the answer are processed through your existing coding-agent provider account. Valv does not proxy them through a hosted Valv service or require a separate API key.
Ask your database a question.
Free for personal use, every feature included. Commercial use needs a seat at beta pricing. See pricing.