Owner Assistant — query in natural language
SaseKY AI accesses your data directly, calls one of seven built-in tools when needed, and summarizes the answer in your sector's language.
Secure architecture: read-only queries, five-second timeout, one-thousand-row result cap and multi-layer security audit. Write operations require approval from the company owner.
You
Who's my top technician this month?
SaseKY AI
Mehmet Y. leads with 12 jobs this month. Total revenue contribution ₺52,400.
You
Did our profit grow vs last month?
SaseKY AI
Last month net profit was ₺72,800. So far this month ₺85,200 — 17% growth.
You
What products are running low?
SaseKY AI
5 products are critical: Motor Oil 5W30 (3 units), Brake Pads (2)…
What the AI does today
Live capabilities of SaseKY AI — speaking your sector's language with direct access to your data.
Chat & sessions
Context-preserving dialog
Remembers prior messages and understands follow-ups. Stays on topic through long multi-turn conversations.
Unlimited sessions
Open as many chats as you want, name them, pin or archive. Switch from the side menu with one tap.
Streaming response
Answer streams word by word. No long waits, even on lengthy replies.
Suggested starter prompts
Empty state shows four ready prompts: invoices, VAT, work orders, best sellers. One tap to send.
Data access
Seven built-in tools
Customer, balance, invoices, work orders, stock, regulation and web search. The assistant picks the right one for your question.
Multi-step reasoning
On complex questions the assistant chains multiple tools, reads intermediate results, then composes the final answer. You can review the steps below the response.
Write requires approval
Operations that change data require push-notification approval from the company owner. No record is updated without explicit approval.
Up-to-date information lookup
For regulation or current information the assistant queries trusted web sources; references appear beneath the answer.
Natural-language reporting
Schema-validated output
Every generated query passes schema validation; the chance of broken or hallucinated reports is greatly reduced.
Read-only with hard limits
Data-modifying commands are blocked; queries over five seconds are aborted; results over one thousand rows are truncated.
Multi-layer security audit
Comment stripping, sensitive-field filter and parametric binding. Every query passes automated audit before execution.
Sector terminology dictionary
Speaks each sector's language. Terms like "technician", "asset" or "work order" map to the correct data behind the scenes.
Security & KVKK
Role-based access
Natural-language queries are active only for management-level roles. Supervisors cannot access financial data.
Anonymous learning
The global learning pool stores no company ID, customer name or user information. Only anonymous counts and query patterns are kept.
Sensitive-field filter
National ID, password and session keys are automatically stripped from results and from the model's context.
Bilingual
Turkish or English — replies follow the user's language preference.
Answer format
Rich text
Lists, bold/italic and code blocks supported; long answers stay readable.
Table + CSV download
Data-driven answers appear as tables; one tap to download as CSV.
Cache savings
If the answer comes from cache the message shows a badge. Cost on frequently repeated queries is up to 75% lower.
Listen
A "Listen" button below the answer reads it aloud in Turkish or English.
Feedback & learning
Thumbs up / down
Each answer can receive a positive or negative rating. The assistant gradually learns sector patterns.
User correction
When you receive a wrong answer you can write the correct one. It is anonymized and added to the global learning pool; future queries become more consistent.
Successful patterns
Working query patterns are collected sector-wise; on similar questions the assistant draws on them.
Document base infrastructure
Infrastructure is ready for regulation and sector-document search; activates as content coverage grows.
Token system
Transparent token spend
Each query consumes tokens from your balance based on usage. Balance is verified before the request starts.
Bonus-first usage
If a gift or promotional pack exists it is used first, then the main balance.
Auto-refund on failure
Tool timeout or model error triggers an automatic token refund.
Mobile purchase
Token packs available on the mobile app via Google Play.
Infrastructure & performance
- DeepSeek V3.2 — chat, structured output and tool calling
- Cache support — up to 75% cost reduction on frequent queries
- Cloudflare Workers infrastructure — global edge distribution, low latency
- Per-company isolated database — tenant data never crosses tenants
- Encrypted authentication — 24-hour secure session