THE AI-DRIVEN BI ANALYST
Ask your data.
Get the answer,
not a ticket.
For regulated enterprises in banking, insurance, telco and manufacturing that need answers from live operational databases, running on-prem, air-gapped, or in your own cloud under your own LLM keys.
Asky is conversational BI that stays inside your building. Type a question in plain language. Asky finds the right table, generates safe SQL, dry-runs it, and returns a chart with the business read.
Three ways in. Read the architecture first if you would rather not book anything yet.
On-prem & air-gapped/BYOK LLM keys/EXPLAIN safeguard/Oracle · PostgreSQL · MySQL · MSSQL
Illustrative demo data: synthetic tables, not customer results. Predictive scores shown here are produced by B2Metric ML; Asky’s own scope is the governed query and the answer on top of them.
See it work
Don’t read dashboards. Ask Asky.
Watch Asky take a plain-language question, find the right table, guard the SQL and hand back a chart with the business read.
In production
A telecom operator, on their own infrastructure
Deployed inside their own environment
“At stc Kuwait, our CLM & Business Intelligence team used to wait couple of hours and days every time a business question needed a new report team ticket. With Asky, our team can ask churn or campaign performance questions in plain language and get an answer with a chart in seconds, without a ticket to the data team.”
“What mattered most for us was that Asky runs on-prem/inside our own environment as a telecom operator, data residency made that non-negotiable. We also expect a reduction in ad-hoc reporting requests, faster time-to-answer on churn segments, campaign segments built end-to-end.”
Churn and campaign questions answered without a report ticket.
Runs inside the operator’s own environment.
From question to audience, inside the same governance path.
The data-team bottleneck
Business users wait days for answers. Every new question becomes an IT ticket, and every dashboard rots right after the one question it was built to answer.
The queue is the product
A question that takes a minute to ask takes days to answer, because it has to pass through the one team that can write the SQL.
Dashboards answer yesterday
Each one is built for a single question. The follow-up question, the one that actually decides something, starts a new ticket.
Self-service stops at the perimeter
The tools that make asking easy assume the data can move to their cloud. In a regulated environment, that is where the project stops.
We keep market statistics off this page unless we can source them. Measured pilot results (accuracy, time-to-answer, deflected requests) are shared with their methodology during evaluation.
How it works
Question in, governed answer out
Ask in plain language
Any language, in the words your business already uses. No SQL, no dashboard hunting.
Find the right table
Mem0 semantic memory maps your schema and business rules, learned from conversation, not from a rebuild.
Guard, then execute
Rule-based engine first, LLM only as fallback. Read-only SQL, EXPLAIN dry-run and cost estimate before a query ever runs.
Answer + chart + insight
A chart and a plain-language business read, not a SQL string handed back to a non-technical user.
Why Asky: three pillars
Your data stays yours
On-prem, BYOK LLM, CMEK. No hyperscaler dependency, the default choice for regulated industries.
Memory that learns
Tell Asky once that revenue is net of refunds and it holds that definition for you and your team, reviewed and approved by a human before it becomes a team convention. No ticket, no model rebuild.
Safe by default
EXPLAIN pre-execution guard, read-only SQL, 3-layer token quotas and full audit logs.
What only Asky does
EXPLAIN safeguard
Every query is read-only, dry-run through EXPLAIN and scored for estimated rows, cost and runtime. Anything over the thresholds your DBA sets is blocked, not queued.
True air-gapped on-prem
In the local deployment, query and result stay inside your network with no outbound call, including with a local model. The cloud BYOK path is documented separately below.
3-layer token quotas
Finance controls LLM spend at session, user and team level, not just a monthly cap.
Mem0 semantic memory
Learns “revenue = net revenue − refunds” from a single chat message, scoped to a user or a team. Implementation detail lives in the docs, not in the sales pitch.
BYOK LLM per customer
Bring your own OpenAI or Azure key, or a fully local model. Asky never stores your credentials.
Per-customer language
Every customer asks and reads answers in their own language, any language, with the same guardrails applied.
Asky vs. the market
Where conversational BI stops being available to a regulated enterprise.
| Capability | Asky | ThoughtSpot | Databricks Genie | Snowflake Cortex | Wren AI |
|---|---|---|---|---|---|
| On-prem / air-gapped | Yes | No | No | No | Cloud only (paid) |
| BYOK LLM keys | Per-customer | Experimental | No | No | No |
| Multi-DB (Oracle / PG / MySQL / MSSQL) | Yes | Cloud-only | Lakehouse only | Snowflake only | Connectors |
| EXPLAIN safeguard | Built in | No | No | No | No |
| 3-layer token quotas | Yes | No | Compute billing | Per-message | No |
| Semantic memory | Learns | Static graph | Ontology | Stateless | No |
| On-prem governance (RBAC / RLS / CLS / audit) | Full | Cloud only | Unity (cloud) | Cloud only | Paid cloud only |
Assessed for a self-managed, air-gapped deployment of each product against public vendor documentation, last verified August 2026. “No” means the capability is not available in that deployment mode; several of these products offer it in their managed cloud. Pre-execution dry-run and cost estimation exist in other environments; what we claim here is the mechanism and its thresholds, not novelty. Row-level sources are in the architecture brief and we will walk through any row you disagree with.
Deployment
Where does the data actually go?
Two deployment paths, two different answers. We describe them separately because they carry different assurances.
Nothing leaves your network
Egress requirement: none. Suitable where an internet route is not permitted at all.
Rows stay put; the prompt does travel
Egress requirement: one allow-listed model endpoint. Choose the local path if that is not acceptable.
Evaluation
30 days, on your data, against your criteria
Three to five tables, one business area. Definitions documented, SSO and roles connected.
Your own question set, scored with your analysts. Access, row and column controls verified.
A measured accuracy result with its methodology, analyst hours saved per question, and an integration plan.
The questions buyers actually ask
Won’t the LLM hallucinate SQL?
Rule-based engine first, LLM only as fallback. Every query is guarded and EXPLAIN dry-run, with self-correction on error. Authorized roles can view and verify the SQL.
Does our data go to an AI outside?
It depends on the deployment path. Local / air-gapped: nothing leaves your network, including the model. Your cloud with BYOK: the question text and schema metadata reach the model endpoint you nominate under your own key and contract, while result rows never do. Both paths run queries in your own warehouse behind SSO + RBAC.
Does everyone see all the data?
SQL visibility is role-based: viewers see only the answer and chart. Multi-tenant isolation keeps user and team scopes one-to-one.
Our schema is custom and keeps changing
Adapted with no code. Update the vault .md files (keywords, order, insights) and Asky follows.
Will LLM cost run away?
3-layer token quotas at session, user and team level, set by admins. Local LLMs are counted too.
We already have Power BI
Asky is a natural-language layer, not a rival. It closes the open-ended questions dashboards can’t answer.
Start small
One warehouse, one business unit,
a few vault topics
A fast, low-risk proof of concept. Success is measured on questions answered without asking the data team, and on time-to-answer.
- ■The demo adapts real business test-case scenarios to your data
- ■You define the success criteria before we connect anything
