Deployment
One product, two deployment paths.
Deployment happens on your own server; the details of both paths follow.
Two paths
Two paths, one product
The choice is made at the source and AI layer; the permission, approval and audit backbone is the same on both paths.
Pilot path
A fast start with cloud AI and your approval. Your source connects with read permission only; row sharing is off by default.
Local model path
With in-house AI, an in-house search model and voice, closed external channels and closed cloud connections, zero external traffic is the goal; it is verified at deployment by measuring outbound network traffic. Deployment alone is not proof.
Journey
The rollout journey
Four steps from discovery to rollout.
0.5 day
01
Discovery
Source, question set, permission model and goals are clarified together.
0.5–1 day
02
Deployment
The deployment is done, the source connects, and capabilities are deliberately switched on.
2–4 weeks
03
Pilot
You work with real questions; golden questions and institutional knowledge mature.
at your pace
04
Rollout
Channel and user scope expand through group-based access.
FAQ
Frequently asked questions
Do we have to replace our existing ERP?
No. Yaverim connects read-only to your SQL Server or Oracle system, or uses your Power BI data models as a source. Write intents go only through approved flows.
Does data go to the cloud?
In text chat, your question, business context and summary go to the cloud AI; rows are shared only with administrator permission and hidden.
Details on the security pageWhich sources are supported?
SQL Server and Oracle databases, your Power BI data models, and documents you upload to chat. One active data source per deployment is the rule.
Can we enforce row-level restrictions?
The rule model and admin screen are ready; definitions are recorded today, and applying them to questions is the next step. Today, an approved list per table, hiding of personal-data fields and summary-only viewing by default are in effect.
Will it be hard to run after the pilot?
The admin screens bring connections, user permissions, channel switches, quotas and model management into one place. The rollout journey is defined as discovery, deployment, pilot and rollout.
Deployment
Installed on your own server
Deployment happens once on your own server, and the AI components run on networks kept apart from each other. One isolated deployment per customer is the rule, and secrets are held in your database. Air-gapped and browser variants are also available.
Technical requirements (for your IT team)
Requirements
The pre-deployment checklist is short and clear.
- 01
Database
MS SQL Server 2019 or 2022; collation Latin1_General_100_CI_AS_SC_UTF8 with RCSI enabled.
- 02
Operating system
Linux container environment. Windows and IIS paths are ready; real-hardware validation is ongoing.
- 03
Source connection
One active data source per deployment: SQL Server, Oracle or a Fabric semantic model. Connects with a read-only account and an approved list.
- 04
Documents and scanning
A shared directory for the document store; a scanning package set up by the operator for reading images and scanned pages.
- 05
Model access
A cloud provider key or an on-premises OpenAI-compatible endpoint. Secrets are held in the application database.
# compose.yml services: migrate: app: worker: vector-sidecar: ollama: mssql:
A half-day pilot and a live demo on your own data
Discovery typically takes half a day, deployment half a day to a day, and the pilot two to four weeks. Fill in the form for a quote, or write to [email protected] directly.