Questions people actually ask first.
Including the two most people ask before connecting anything: can it write to my data, and who else can see my queries.
Which databases does BeyondQueries support?
PostgreSQL and MySQL, with SQL Server and Oracle on the way. You connect a database you already run, and BeyondQueries discovers the schema on connect — you do not import or copy your tables anywhere.
Can BeyondQueries modify or delete my data?
No. Every statement is parsed before it reaches your database, and anything other than a single read-only SELECT is refused: INSERT, UPDATE, DELETE, DROP and ALTER, a second statement tacked onto the first, a write hidden inside a CTE, commands such as COPY, and anything the parser cannot read. Connecting with a read-only database user is still the right habit — it makes the same guarantee at the database itself, as a second line.
Do I need to install anything?
No. BeyondQueries is hosted and runs in the browser. There is no desktop client to download, no local agent, and no per-machine setup — which also means a correction one person makes is available to everyone else immediately.
How does BeyondQueries learn from a wrong answer?
On your own, you write table and column notes — say what a column means or how a figure is calculated, and every later question starts from it. On a team plan, anyone can flag an answer as correct, incorrect or partially correct with a comment. Flagged queries go into a review queue, and someone with review permission approves or rejects the correction. Approved corrections are stored in a vector knowledge base and retrieved as grounding for later questions. Nothing is fine-tuned, so an approval takes effect immediately and you can see what was retrieved.
Who on my team can see which queries?
It depends on their role. Owners and admins can see every query and all analytics for an instance. Editors see all analytics but only their own queries. Analysts and plain users see only their own. Roles are granted per database instance, so someone can be an admin on a warehouse and a plain user on production.
Can I use my own model API key?
Yes. An instance can carry its own LLM API key, which is encrypted at rest with Fernet and never returned to the browser after it is saved. Most of the per-question cost sits in that key, so bringing your own is also the cheapest way to run BeyondQueries.
Is my database password stored in plain text?
No. Connection strings, database passwords and LLM API keys are encrypted at rest with Fernet — AES-128 in CBC mode with HMAC authentication — and are never shown again after they are saved.
Can I test schema changes before they affect everyone?
Yes. Table and column instructions and connection settings can be exercised in a sandbox copy of an instance and then promoted to production once they behave. Only owners and admins can promote a sandbox.
What does the model see of my data?
Your question, the tables and columns you left visible, your notes and approved corrections, and rows from the result so it can write the answer in words. Tables and columns you switch off are left out of its context and refused by the parser even if a query names them. Database passwords never reach it. You choose the model, including one you run yourself with Ollama or vLLM.
Was BeyondQueries called DataSpeak?
Yes. The product was called DataSpeak until September 2026, and only the name changed — accounts, connected databases, notes and approved corrections carried over as they were. It is not related to the other products that share the old name, such as a desktop database client and a data-science consultancy.
What happens to my corrections if I stop using BeyondQueries?
They belong to your instance, alongside the table and column instructions your team wrote. Export is not yet a one-click feature — if you need your knowledge base out, ask Irevex and it will be provided rather than held hostage.
Ask, and this page gets longer.
If your question is not above, it is worth adding. Ask Irevex, or read how the correction loop works and what the role matrix allows.
The fastest answer is a connected database.
Point BeyondQueries at something non-critical with a read-only user and ask it three real questions.