Skip to content
Get a quote in 24 hours

ActiBot AI Labمختبر أكتي بوت للذكاء الاصطناعي

AI trained for the work your organisation actually does. We take an open local model, teach it your skills, your terms and your language, test it against an exam it has never seen, and ship it ready to run on your own servers, connected to K2 and the systems you already use.

Trained on your use casesRuns on your serversConnects to K2 and your systemsArabic dialects and EnglishCPU or GPUNo outside AI calls
ActiBot AI Lab logo
Round 4 · check passed
Sealed exam passed
Runs on premises
26AI skills ready to train
AR · ENGulf dialects and English
CPU · GPUWindows or Linux servers
0calls to outside AI services at run time

A complete AI product, not a model file

ActiBot AI Lab is a product in its own right. It works beside K2 or any of your existing systems. Each product holds everything needed to run the AI in your environment and plug it into the systems your people already use.

Your trained model

An open local model, trained on your skills, your vocabulary and your examples. It answers under the ActiBot AI name or your own product name.

  • Skills chosen from your use cases
  • Arabic, dialects and English
  • Your terms, tone and formats

Runtime package

A signed installer that sets the AI up as a service on your Windows or Linux server, on CPU or GPU, sized for your users.

  • Runs as a background service
  • One internal address and an access key
  • Checksums and a signed record

Connections

Everything needed to call the AI from your systems, with K2 ready on day one.

  • K2 service instance, ready to use in forms and workflows
  • Standard API description for any other system
  • Built into Masar and Sejil

Evidence and care

A test report you can read, a model card that says what the AI does and does not do, and a path to better versions.

  • Exam results per skill
  • Safety and leak test results
  • Optional ongoing improvement, from usage codes, never your content

From your use case to a tested AI, in one guided journey

Every model follows the same journey. We agree what "good" means before training starts, keep a final exam sealed until the end, and release only when every target is met and a person has signed off.

Optional ongoing improvement: lessons from real use, mistake codes and counts only, never your content 1 Define skillsand targets 2 Prepare thematerial 3 Measure thestarting point 4 Training loop Final examand sign-off 6 Packageand sign 7 Install andconnect In use REPEAT UNTIL TARGETS ARE MET A · Teach B · Check C · Study the mistakes D · Write better examples E · Tune Targets per skill,agreed with you Cleaned, de-identified,exam sealed away Runs on yourhardware, signed K2, Masar, Sejiland your systems Release gates ✓ Passes its sealed exam✓ No skill worse than before✓ Safety and leak tests✓ Answers with its own name✓ A person signs off
Mint is the model's path from your use case to your server. The gold path is optional. For customers who want ActiBot to keep improving their model, lessons from real use come back as targets for the next version, and each new release must beat the last.
Before training

We agree what good looks like

Your use cases become a list of skills. Each skill gets an example, a measurable target and a set of exam questions you can review.

Skills, targets and an exam, approved by you
During training

The model learns from its mistakes

Every round, the model is checked. Its mistakes are studied, new examples are written for exactly those gaps, and it trains again.

Progress visible round by round
Before release

It passes an exam it has never seen

The final exam stays sealed until the end. A version ships only when it passes, keeps every earlier skill, and a person signs off.

An evidence report with every release

Your knowledge in, your secrets kept

Most of the effort in a good model is in the material it learns from. We turn your documents, rules and examples into clean training material, and we keep anything confidential away from outside services.

Your materialdocuments, rules, samples Clean and de-identifynames and numbers masked Add invented examplesevery case, every dialect A person reviewswrong rows removed Training materialwhat the model learns from Sealed final examnever seen in training Material marked confidential is never sent to any outside AI service. Outside help, when approved, sees only invented examples.

Four use cases, live

Each view shows a model trained by ActiBot AI Lab at work. The AI suggests and explains; your people decide. Choose a use case.

AIRecruitment · AI interviewLiveSample data
REC 03:42
Question 3 of 6
في مشروعي السابق تأخر المورد أسبوعين، فقسمت التسليم إلى مرحلتين واتفقت مع العميل على أولويات واضحة، وسلمنا الجزء الأهم في موعده.
Competency · Problem solvingAnswer in Arabic · Gulf dialect

Tell us about a time a supplier delay put a deadline at risk. What did you do, and what was the result?

Translation for the panel In my last project the supplier was two weeks late, so I split the delivery into two phases, agreed clear priorities with the client, and we delivered the most important part on time.

Answer evaluationStrong · 4 of 5
Relevance to the question92
Evidence of the competency86
Structure (situation, action, result)88
Language and clarity81
Evidence quoted“split the delivery into two phases … delivered the most important part on time”
✓ On topic✓ Spoken answer, not read outRecruiter confirms the score
Shortlist ranking · Project coordinatorUpdated after each interview
#CandidateCV matchInterviewOverall
1Candidate C-0418 ▲ 288%4.4
2Candidate C-039291%4.0
3Candidate C-040784%3.9
4Candidate C-037179%3.4

26 skills to build your AI from

We map your use cases to the skills below, then train only what you need. One model can hold several skills, and new skills can be added in the next version.

Documents

  • Pull fields out of documents
  • Classify documents
  • Summarise
  • Read tables
  • Read scanned pages and images
  • Spot personal data

Language

  • Draft replies and letters
  • Translate Arabic and English
  • Write reports
  • Read tone and sentiment
  • Bilingual assistant
  • Answer from your documents

People and hiring

  • Match CVs to jobs
  • Write job descriptions
  • Score against a rubric
  • AI interviewer

Search and decisions

  • Suggest where work goes
  • Flag risks and anomalies
  • Smart search and duplicates
  • Questions to reports

Speech and vision

  • Speech to text, Arabic dialects
  • Text to speech
  • Check video frames

Actions

  • Design workflows from text
  • Fill forms from requests
  • Take actions, with your approval

On your servers, connected to your systems

Training happens at ActiBot. What runs at your site is a signed package on your own hardware, inside your network. Your systems call it through one internal address, and no question or answer leaves your environment.

YOUR ENVIRONMENT · ON PREMISES OR PRIVATE CLOUD ActiBot AI service Your trained model CPU or GPU · Windows or Linux one address · access key K2 forms and workflows Masar Sejil Your other systems Staff and portals K2 connects through a ready service instance; other systems through a standard API description ActiBot AI Lab training and new versions signed package optional:usage codes only No calls to outside AI services
Mint lines are calls that stay inside your network. The gold line is the signed package we deliver; new versions arrive the same way. The dotted mint line is optional: it is switched on only for customers who want ActiBot to keep improving their model, and it carries mistake codes and counts, never your content.

Every version ships with its exam results

You see what the AI was trained to do, how it scored on each skill against the agreed target, and how it compares with the version before.

Release report · Version 1.3✓ Ready to shipSample data
SkillTargetBeforeThis version
Pull fields out of letters9588.196.4Passed
Suggest where work goes9084.092.7Passed
Score interview answers9086.591.2Passed
Arabic dialect speech to text9089.090.8Passed
Safety and leak tests9596.098.3Passed
✓Sealed exam passed for every skill
✓No skill scored lower than before
✓Answers with its own name in Arabic and English
✓Never reveals its instructions or another customer's data
✓Signed off by a person

Speaks the way your people speak

Arabic first

Trained on Modern Standard Arabic and the Gulf dialects your users speak, including spoken answers, with English alongside.

Your vocabulary

Your sector's terms, your document types, your reference formats and the Hijri dates in your letters.

Your rules

What the AI may suggest, what it must never decide, and when it hands over to a person, all trained in and tested.

The AI behind Masar and Sejil

Our K2 add-ons use models from ActiBot AI Lab. The same AI service can serve them and your own K2 processes at once.

Masar logo

Masar Process Suite

Turns a written procedure into steps, roles, decisions and approvals, ready to map and refine.

Sejil logo

Sejil

Reads incoming letters, suggests where they go, summarises, spots personal data and drafts replies.

Why organisations choose a trained, private AI

01

Your data stays yours

The AI runs inside your network. Questions, documents and answers never travel to an outside AI service.

02

Built for your tasks

A model focused on your exact work, sized for speed on the least hardware you need to control, with no per-question fees to an outside service.

03

Proven before it ships

Targets agreed with you, a sealed exam, safety tests and a signed-off evidence report for every version.

04

Ready for your systems

K2 connects on day one. Masar and Sejil use it built in. Anything else connects through a standard API.

05

Runs on the hardware you have

We size the model to your servers, CPU-only or GPU, and tell you up front how fast it will answer.

06

Gets better every version

Optional: if you want us to keep improving your model, usage codes from your site become targets for the next round, and each version must beat the last. Without it, your model stays exactly as delivered.

From first call to your AI in production

Step 1DiscoveryTell us your use cases or share your requirements. We map them to skills and targets.
Step 2Proof on samplesWe measure a starting model on your redacted samples, so you see the gap before you commit.
Step 3Train and sign offTraining rounds with visible progress, then the sealed exam and your sign-off.
Step 4Install and improveWe install it and connect your platforms and systems. If you choose ongoing improvement, improved versions arrive through our after-sales service.

Tell us the AI skills your teams need

Book a discovery session. Bring one use case and a few redacted samples, and we will show what a trained model could do with them.