Login AI
Corporate AI assistant for your knowledge base
We build a corporate assistant over your knowledge sources: one search index, permissions from the corporate role directory, and a citation from the retrieved fragment in every answer. The contour runs on your hardware or in a closed cloud, and request logs stay with you.
The assistant finds answers in email, files, the wiki, and databases and links to the source.
Pricing
from ₽1,200,000
per project
This is the lower bound of a project. We scope the exact quote after a brief: it depends on volume and the number of integrations.
Capabilities
One index across sources
Email, files, the wiki, and databases land in one search index. Each answer shows which system the fragment came from.
Roles from the directory
The assistant reads groups and roles from the corporate directory and shows an employee only the documents open to them.
A citation in every answer
An answer relies on the retrieved fragments, and every claim carries a link to the document, section, and source version.
A closed contour
The assistant runs on your hardware or in a closed cloud. The request log keeps the question, the retrieved fragments, and the response time.
From request to result
1
Source inventory
We list the systems, document volume, and owner of each source. We fix which data enters the index and who keeps it current.
2
Access model
We review groups and roles in the corporate directory, set inheritance rules, and design an index schema with access metadata on every fragment.
3
Search and answers
We bring up hybrid search, connect the model through vLLM, and build the answer layer: retrieved fragments, citations, and an access check before delivery.
4
Pilot and launch
Nine weeks to the first contour: one team, three sources, typical questions. We measure the share of answers with a correct citation and move the contour into production.
Check if it fits you
Good fit
Knowledge spread across systems
Email, files, the wiki, and databases live apart. The assistant brings them into one search and shows the source of every answer.
A corporate role directory
Groups and roles already live in the directory, so access inherits automatically, and a new employee gets the right documents on day one.
Not a fit
Sources without owners
No one is responsible for document freshness. In this case the project starts with named owners and an update policy.
A public-facing chatbot
The assistant answers employees inside the perimeter. Customer support and sales chat belongs to a separate service with a different scenario and access model.
What you get
1 contour
The first contour
The first contour brings email, files, and the team knowledge base into one search. The team gets answers with links and works in the systems it already uses.
100–1,000
Users and roles
The calculator covers 100 to 1,000 users and accounts for role permissions. Cost grows with the number of employees and sources.
9 wks
Time to launch
The source audit, search build, and launch of the first contour take nine weeks. The pilot runs with one team, then we connect the remaining groups.
What the delivery includes
Source map
The map shows the system, owner, volume, and update frequency of each source. It names the systems we connect first.
Search with access control
Search runs across all sources and checks directory permissions before delivery. Every answer carries a link to the document and the fragment version.
Pilot and training
The pilot runs with one team and three sources. Training shows employees how to ask a question, read a citation, and report a doubtful answer.
How the system is built
Knowledge index
Email, files, the wiki, and databases pass through connectors into one index with metadata for source, version, and access. Search finds a fragment by meaning and exact identifier, so an answer rests on a verified document.
Role-based access
The assistant reads groups and roles from the corporate directory and checks permissions on every fragment before it builds an answer. An employee sees documents for their group, and the log records the access basis for each request.
Deployment inside the perimeter
The contour runs on your hardware or in a closed cloud, so data stays inside the company perimeter. Request logs and indexes stay on the agreed site, and model updates pass through readiness checks.
Where the solution pushes back
- 01
Answer quality depends on sources
Clean sources with an owner and a version produce an accurate answer with a link. The source map and update policy set that level before the start.
- 02
The role directory stays current
Access inherits from the corporate directory, so current groups and roles produce an accurate result. Directory maintenance is part of the operations policy.
- 03
On-prem adds an infrastructure stage
Deployment on your hardware adds a stage for site preparation, replicas, and monitoring. That stage is part of the project plan and the budget.
Technologies
The technologies we use in our work and their role in a project.
Search, access, and deployment
OpenSearch holds the full-text and vector index, Keycloak checks roles on every request, PostgreSQL stores the indexing queue and logs, Kubernetes runs the contour across sites, and vLLM serves the open model.
Cost calculator
This is an indicative range, not a quote. We scope the exact estimate after a brief: it depends on volume and the number of integrations.
Users
Knowledge sources
Access control
On-prem contour
Add-ons
Support
Year one: updates and monitoring, a reply within 2 business days.
Service
Corporate AI assistant for your knowledge base
The estimate covers our team's work. Hardware, third-party service licences and travel are added on request.
Estimated cost
₽1.7M
Timeline
9 wk
Frequently asked questions
Which sources can we connect?
Email, files, the wiki, databases, and internal portals. We configure a connector for each system during the audit and fix the source list in the map.
How does document access work?
The assistant reads roles from the corporate directory and checks access before returning a fragment. An employee sees answers from documents open to their group.
Where does the data live?
In your contour or in a closed cloud with an isolated network. Request logs, the index, and the model stay on the agreed site.
How long does the pilot take?
Nine weeks to the first contour: source audit, search build, and launch with one team. After that we connect new sources and groups.
What teams add to this service
Related Services
Part of Solutions
Implementation Examples
−70%
Request handling time
Ready to discuss your task?
Leave a request – we'll get back with a proposal and a preliminary estimate.
Contact us