FLUXL فارسی

On-premises · Bilingual · Secure · Measurable

Your organization's knowledge, one question away

FLUXL turns your documents, procedures and technical manuals into precise, cited answers, in Persian or English, entirely on your own infrastructure.

Request a pilotExplore the products

The challenge, and how FLUXL answers it

ChallengeFLUXL's answer

Scattered knowledge

Procedures, policies, technical manuals and team know-how are spread across thousands of files

One assistant that reads them all and answers with the exact document and section

Data confidentiality

Internal documents and customer data must not be sent to a foreign cloud service

Fully installed on your servers, even on an isolated network with no internet

Persian language

General-purpose tools understand Persian poorly and miss your domain terms

Persian interface, answers and voice input; Persian questions are translated to search English documents

Speed and quality

The right answer depends on a few people, and onboarding takes months

Answers in seconds, with search quality measured automatically against reference questions

One platform, every channel and knowledge source

Enterprise documentsAutomatic syncWeb search and readingWeb app and voiceBale messengerCoding assistantOn-prem GPU serversSSO and access control

  1. Inputs

    Word, Excel, PowerPoint, PDF and scanned images with Persian OCR; automatic sync from Nextcloud and Git

  2. Core

    Hybrid keyword and semantic search, access control, and answers with cited sources

  3. Channels

    Persian web app, voice input, Bale bot, an API for your systems and for coding tools

  4. Infrastructure

    Language models on on-prem GPUs; single sign-on with Active Directory

The FLUXL product family

Knowledge Assistant

Cited answers in Persian or English from each unit's documents; a separate knowledge base per domain, access down to each document

Coding Assistant

Works with tools like opencode; knowledge and web search run on the server, so developer machines need no internet

Meeting Intelligence

A meeting recording becomes a Persian transcript, minutes, action items and a summary, on internal servers

General Assistant

Chat with attached files, projects and personal memory; export to Word, Excel and PDF

QA and Test Center

Test cases designed from requirements, a traceability matrix, CI failure triage and Jira integration

Integration

API and access tokens, a Bale bot, sync with Nextcloud and Git, single sign-on

Use cases across the organization

Support and contact center

Consistent, cited answers to repeated questions from customers and agents

Legal and compliance

Find the relevant clause in contracts, regulations and internal policies, with exact references

IT and operations

Troubleshooting guides, operating procedures and system and infrastructure documentation

Software development

A coding assistant that knows your internal documentation, with no code leaving the network

HR and training

Answers on policies for employees, and fast onboarding for new staff

Management and meetings

Automatic minutes, and summaries of long documents and reports for decision makers

Context enrichment: every answer draws on seven sources

All sources share one context budget, and each is listed in the answer's sources.

  • Enterprise knowledge base

    Hybrid keyword and semantic search, with Persian questions translated

  • Attached files

    Up to 20 files per chat, each given a fair share of the context

  • Project files

    Files attached to the other chats in the same project

  • Personal file memory

    Files the user chose to use in all their chats

  • Links in the question

    Safely reads up to 3 web pages, checking every redirect

  • Live web search

    For the general assistant, one click away

  • Conversation memory

    Rewritten follow-ups and each user's long-term memory

Coding Assistant: knowledge where developers work

Developer

opencode and other OpenAI-compatible tools

FLUXL gateway

Authentication, credits, automatic model choice, policies

Language model

On-prem GPUs; cloud models only if allowed

OpenAI-compatible

Works with opencode and similar tools, unchanged

Smart model choice

By each model's health, speed and context size

Full audit log

Every server-side tool call is recorded and limited

No internet needed

Tools run on the server; results go into the model's context

FLUXL's five-layer architecture

  1. 1. User layer

    Persian and English web app, voice input, Bale bot, meeting intelligence, API

  2. 2. Intelligence and search

    Question rewriting and translation, hybrid keyword and semantic search, cited answers

  3. 3. Knowledge layer

    A separate knowledge base per unit, 15+ document formats with OCR, automatic sync

  4. 4. Model layer

    Open models on your on-prem GPUs; routing by each model's health and speed

  5. 5. Governance and operations

    Single sign-on, document-level access, usage credits, audit log, monitoring and alerts

Enterprise-grade security and confidentiality

Data sovereignty

Data and models stay on your servers; an offline edition for isolated networks, with no outside connection

Access control

Single sign-on with Active Directory and LDAP; separate access per unit and per document

Data masking

Card numbers, IBANs, national IDs and phone numbers, validated; passwords and keys stripped from documents

Audit and traceability

Admin and security audit logs; per-user usage; every answer fully traceable

Reliability

Continuous model monitoring, alerts on failure and slowdown, updates without dropping requests

Testing and quality

Independent penetration test with every finding fixed; thousands of automated tests, a regression test for every bug

Scaling: every layer grows independently and horizontally

  1. Entry and load balancing

    One address for users; requests spread across every service instance

  2. User services

    Stateless; instances scale up and down automatically with load

  3. Question workers

    A shared work queue; adding workers raises concurrent capacity

  4. Data layer

    Vector, relational and cache stores kept separate; clusterable and replicable

  5. Model layer

    Multiple GPU servers and sites; automatic routing by each model's health and speed

A dashed cube is the next instance, added without changing any other layer.

Install on any platform, scale out as you grow

Bare metal

One-script install

Virtual machine

Any hypervisor

Containers

Docker Compose

Kubernetes

Autoscaling

OpenShift

Runs without root

Horizontal scaling: each layer grows on its own

One serverLoad-balanced instancesAutoscaling cluster

Offline edition

All images and local models, for networks with no internet

Model-vendor independent

Any open model or OpenAI-compatible gateway, swappable

Offline signed license

No connection to the vendor's server required

In production, with measured results

383K

Indexed text chunks from the documents of 10 specialist domains

66

Monthly active users, across technical and operations units

21K

Requests a month, knowledge questions and the coding assistant combined

9×

Growth in weekly requests in one month, with no loss of speed

8.3 s

Median full-answer time, down from 15.6 s in the same month

43%

Of questions asked in Persian; answered in Persian, with sources

How we work together

Technical session

Week one

Understand needs, choose the pilot unit, define success metrics

Pilot

30 days

Install on your infrastructure, load one unit's documents, measure with real questions

Rollout

After the pilot

Connect to your single sign-on, add more units, train users

Growth and support

Ongoing

New modules, version updates, support and monthly quality measurement

Why FLUXL

  1. 1

    Your data stays with you

    On-premises and offline, with no dependence on a foreign cloud service

  2. 2

    Persian from the ground up

    Interface, answers, search and voice input built for Persian

  3. 3

    Proven in production

    Running today, with quality measured every week

Next step: a 30-day pilot on one unit's documents