PRODUCT_01 ASKBASE AVAILABLE_NOW
Ask a question. Get an answer you can check.
AskBase answers from the documents your team already has, in Slack or in the app, with numbered citations you can open. Access is filtered inside the query, so two people asking the same question get answers built only from what each of them is allowed to see.
What it does, and where each part stands
-
Cited answers shipped
One answer, with numbered markers that resolve to real documents and a footer saying where it looked. Not a list of links to read yourself.
-
Permission-aware retrieval shipped
Access is filtered in the SQL join against your existing groups, never in Python after the fact, and every answer reports how many matches your permissions removed.
-
Fabricated citations stripped shipped
A deterministic pass remaps every citation marker to a real source and deletes any that points nowhere. An invented reference cannot reach the screen.
-
Four honest statuses shipped
Answered, nothing relevant found, knowledge base empty, permission denied. Four separate paths with four different screens. A refusal is never dressed up as an answer.
-
Schema curation shipped
A curator accepts, edits or rejects the proposed column descriptions, PII flags, synonyms and metric definitions. Every commit is audit-logged. This is the work that makes the answers right.
-
Text-to-SQL over your database in build
Questions answered from a live read-only database. The connector and planner run against a real Postgres in our test tier; it stays off per organization until it is proven against your schema.
This is probably you
- Two or three people answer the same questions every week, and everyone knows exactly who they are.
- The answer exists (in Confluence, in Drive, in a thread from 2023), and nobody can find it.
- A new hire’s first month is mostly interrupting other people.
- Not everyone can see everything, so a single shared search index was never an option.
Shapes it has been configured into
Wrappers of the same engine, not separate products. Where a line carries a figure, the figure is a design target from the scenario named beside it, not a customer result.
- DATA QUESTIONS Drop in your invoice spreadsheet. Ask what you spent on shipping last quarter. The total is computed, not guessed.
- ONBOARDING New hires find their own answers instead of interrupting the person who knows.
- POLICY DESK Legal questions answered at 2am. With citations.
Two numbers we measured. Two we are aiming at.
Measured figures name the run that produced them and come from a review packet in the product’s own repository. Design targets name the scenario they belong to. There is no third kind of number on this page.
What it costs
$499/mo + $40 per seat
Fixed base, per-seat on top, and metered usage above the included quotas. Set 27 Apr 2026 and unchanged since. Configuration for your use case is quoted separately after the assessment.
What we will not claim
It refuses, and it says what it searched. Over your documents there are four separate statuses (answered, nothing relevant found, knowledge base empty, permission denied), and a miss reports how many documents it looked through instead of returning a plausible guess. An answer it does return is graded high, medium or low; it is never silently withheld. The 0.55 refusal floor is the text-to-SQL engine’s, and text-to-SQL is off by default on every organization until it is proven against your schema. We have no published accuracy number yet: the golden-pair harness exists, the scored runs do not.
Thirty minutes tells you whether AskBase fits.
A fixed agenda, a one-page map back within two business days, and no pitch. If AskBase is the wrong tool for your problem, that is what the map will say.
Book Free Assessment