Seekers Core
Seekers Core: a private LLM platform that runs inside your environment.
Seekers Core is the private AI stack we use to build every client system. It connects to your documents and systems, searches and answers in Arabic and English with cited sources, reasons with private language models, and runs agents that complete work, all under one governance layer. It runs on your infrastructure or in approved GCC cloud regions, so your data stays with you.
Definition
What is Seekers Core?
Seekers Core is a private LLM platform: a set of tested components for connecting data, retrieving and citing documents, running language models, and coordinating AI agents, deployed inside your environment. Each client gets its own instance. We configure it for your systems, languages, and rules, then build your use cases on top of it.
Why build every project on one reusable stack?
Most of the work in an enterprise AI project repeats from one project to the next: connecting to systems, indexing documents, handling Arabic text, controlling access, logging, and monitoring.
Read moreShow less
Building all of that from scratch for each use case is slow, and it leaves you with several separate systems to maintain.
With Seekers Core, most of each project reuses components we have already built and tested. The project time goes into what is specific to you: your documents, your workflows, your approval rules, and your evaluation set. That is why a pilot can reach production in 4–8 weeks.
The benefit grows with each use case. Your second and third use cases run on the same connectors, governance, and monitoring as the first, so each one is faster to add and easier to support.
Where can Seekers Core be deployed, and which models does it use?
Seekers Core runs wherever your data rules allow.
- On-premiseData lives: Your own data centre
Best when: Strictest data rules, or air-gapped systems
- Private cloudData lives: Your dedicated cloud tenancy
Best when: You already run on a private cloud
- Sovereign GCC cloudData lives: An approved in-country cloud region
Best when: You want speed without buying hardware
Read moreShow less
We size the deployment during the Readiness Sprint and agree it with your IT and security teams.
For models, we choose from open and commercial models that can run privately, based on your language needs, accuracy targets, and hardware. No part of the stack depends on a single model vendor, so you can change models later without rebuilding.
- On-premise: in your own data center, on your hardware, with no outside connection required.
- Private cloud: in your organisation's private cloud, under your network and security controls.
- Sovereign GCC cloud: in approved cloud regions inside the GCC, for organisations that need local data residency without running their own hardware.
How does Seekers Core integrate with our existing systems?
We integrate with what you already run.
Read moreShow less
You do not move your data into a new platform. Connectors use each system's existing interfaces and the access rules it already enforces.
We agree each integration with your IT team: which systems are in scope, what Seekers Core may read, what it may write, and through which service accounts. Write access is added only where a use case needs it, and sensitive writes go through human approval.
Some older internal applications have no usable interface. For those, we look at options such as scheduled exports or a controlled document folder, and we record the trade-offs in the architecture so your team can weigh them.
How is Seekers Core operated after go-live?
After go-live, we run the platform for you as part of Run & Scale, or we support your own team to run it.
- AI draftsReply, memo, or update
- Human approvesThe right role signs off
- Action runsIn your systems
- LoggedWho, what, when
Read moreShow less
The same checks apply either way.
- Monitoring of availability, response times, and usage.
- Accuracy and drift reviews against the evaluation set, repeated as your documents change.
- Model, prompt, and retrieval updates, tested before release.
- Review of audit logs and approval patterns with your risk team.
- A monthly report on usage, accuracy, issues, and planned changes.
Five layers in one stack
◇ Governance wraps every step
- 01ConnectDocuments, ERP, CRM, email
- 02UnderstandArabic + English search, cited
- 03ReasonPrivate models plan the task
- 04ActDraft, file, route, update
- 01
Connect
Reads
- indexed inside your network
- including scanned PDFs
- Arabic text
- +1
Does
Links Seekers Core to the places your information already lives: document stores, shared drives, databases, email, ERP, CRM, and internal systems.
Governed by
Each connector keeps the permissions of its source.
Read moreShow less
- What it does
- Links Seekers Core to the places your information already lives: document stores, shared drives, databases, email, ERP, CRM, and internal systems. Nothing is copied outside your environment.
- How it works
- Connectors read through each system's existing interfaces, using service accounts you approve. Documents are processed and indexed inside your network, including scanned PDFs and Arabic text, and stay in sync as the source changes.
- Governance
- Each connector keeps the permissions of its source. If a user cannot open a file in your document system, the assistant will not show it to them.
- 02
Understand
Reads
- the most relevant passages are retrieved
- mixed Arabic–English text
Does
Searches your documents in Arabic and English and answers questions with cited sources, so users can check the passage behind every answer.
Governed by
Every answer links to the source document and page.
Read moreShow less
- What it does
- Searches your documents in Arabic and English and answers questions with cited sources, so users can check the passage behind every answer.
- How it works
- Retrieval-augmented generation (RAG). The question is matched against your indexed content, the most relevant passages are retrieved, and the model answers from those passages. Search handles Arabic spelling variation and mixed Arabic–English text.
- Governance
- Every answer links to the source document and page. When your documents do not support an answer, the assistant says so instead of guessing.
- 03
Reason
Reads
- We choose models for your data
- languages
- accuracy targets
- +3
Does
Runs private language models that interpret requests, plan multi-step tasks, and explain how they reached an answer.
Governed by
Models run inside your environment.
Read moreShow less
- What it does
- Runs private language models that interpret requests, plan multi-step tasks, and explain how they reached an answer.
- How it works
- We choose models for your data, languages, accuracy targets, and hardware, from open and commercial models that can run privately. Models can be replaced as better ones appear, without rebuilding the use cases on top.
- Governance
- Models run inside your environment. Prompts, model versions, and outputs are logged, so any answer can be traced and reviewed.
- 04
Act
Reads
- Agents follow defined workflows
- use only the tools you approve
- such as opening a ticket
Does
AI agents that complete work across your systems.
Governed by
Sensitive steps stop for human approval.
Read moreShow less
- What it does
- AI agents that complete work across your systems. They draft replies and reports, file documents, route requests, update records, and send summaries.
- How it works
- Agents follow defined workflows and use only the tools you approve, such as opening a ticket, matching an invoice to a purchase order in the ERP, or drafting a letter for review. Each tool has a narrow permission.
- Governance
- Sensitive steps stop for human approval. Nothing is sent, posted, or paid until the named approver confirms it.
- 05
Governance
Reads
- retrieval
- model call
Does
Wraps the other four layers with access control, audit logs, human approval, and monitoring.
Governed by
You set the rules: who can use each use case, which actions need approval, and how long logs are kept.
Read moreShow less
- What it does
- Wraps the other four layers with access control, audit logs, human approval, and monitoring.
- How it works
- Role-based access ties into your identity system. Every query, retrieval, model call, and agent action is written to an audit log. Monitoring tracks accuracy and drift against an evaluation set built from your own documents.
- Governance
- You set the rules: who can use each use case, which actions need approval, and how long logs are kept. Your risk and audit teams can review them at any time.
Questions about Seekers Core
Is Seekers Core software we set up ourselves?
No. Seekers Core is a platform we deploy, configure, and build on for you. Each client gets its own instance inside its own environment. We build your use cases on it and can operate it after go-live, or hand day-to-day running to your team.
Does Seekers Core send our data to an external AI service?
No. By default, models, search indexes, and logs all run inside your environment, and nothing is sent to a public AI service. Seekers never needs a copy of your data on our systems.
Which language models does Seekers Core use?
We select open and commercial models that can run privately on your infrastructure, based on your language needs, accuracy targets, and hardware. You are never locked into a single vendor, and models can be swapped later.
How well does Seekers Core handle Arabic?
We build and test retrieval and agents on Arabic and mixed Arabic–English documents, including scanned files. Accuracy is measured on an evaluation set from your own documents before go-live and reviewed after it.
Can we start with a single use case?
Yes, and most organisations do. The first use case is usually chosen in a Readiness Sprint and built as a pilot. Later use cases reuse the same stack.
See where Seekers Core fits in your environment.
Start with a Readiness Sprint. In 2–3 weeks we map your systems and data, pick the first use case to build on Seekers Core, and give you a costed roadmap.