Everything runs inside your walls.

Three products that work on their own or together. AE-1 puts AI in front of your people, Interstellar connects it to your systems, and Sharq Engine trains the model on your data.

Start a pilotSee each product
AE-1 answering live: meeting notes added with the + button become action items, decisions and an Arabic summary, streamed word by word from a model running on our own hardware. Recorded from the product.

See it working.

Footage recorded from the product, and illustrations drawn in code. Every tile opens the product it shows.

AE-1Type Arabic, English or both; the message box follows you.
AE-1All six Arabic plural forms, counted correctly.
File shareSharePointPostgreSQLS3 archiveSQL ServerPermission checkFinance analyst3 of 5 sourcesHR officer3 of 5 sources
InterstellarEach person only receives what they could already open.دولفينAE-1Designed in Arabic, down to the letterforms.
Open base modelYour model, trained on your records
Sharq EngineA general model, pulled toward your data.
AE-1Four themes, light and dark, switched live.0outbound connections by defaultAE-1Works with the internet switched off.

AE-1

An AI workspace that never leaves your building.

Your staff ask questions, draft and summarise in Arabic, English or both in the same sentence. AE-1 runs as one program on your own servers, and works with no internet connection at all.

Explore AE-1

Arabic, English, or both.

Type in either language, or switch mid-sentence. The message box follows the language you start in.

  • Built for Arabic

    The whole interface reads right to left, mixed Arabic and English stays legible, and numbers and dates follow each person’s settings, Hijri calendar included.

  • Answers from your documents

    Upload PDFs, Word files and spreadsheets and get answers that cite the passage they came from.

  • Your sign-in, your roles

    Staff sign in with the identity provider you already run. Admins choose which models and tools each group can use.

  • Nothing goes out

    No telemetry, no update checks, no third-party calls. Every access is recorded in an audit log your compliance team can export.

Sign-in
Microsoft Entra ID, Okta, Keycloak (OIDC, SAML 2.0, SCIM)

Four themes. Both directions.

Sharq, Nord, One and Gruvbox, each in light and dark, every one checked for readable contrast in Arabic and in English. People pick what suits their eyes; your organisation sets the default.

Interstellar

Connects the assistant to the systems your teams already use.

Interstellar sits between AI and your file shares, databases and business applications. It retrieves what a person is allowed to see, and nothing else.

Explore Interstellar
File shareSharePointPostgreSQLS3 archiveSQL ServerPermission checkFinance analyst3 of 5 sourcesHR officer3 of 5 sources
Illustration. The same question reaches five systems; each person only receives what they could already open.

  • Permissions carried through

    Each person gets results only from what they could already open in the source system. Two people asking the same question can get different answers.

  • Actions need a yes

    Reading is automatic. Anything that changes a record shows exactly what will run and waits for the person to confirm.

  • Credentials stay sealed

    Connection secrets are encrypted at rest and never reach the browser, the model or the logs.

  • A record of every call

    Who asked, what was touched, when, and what happened, in a tamper-evident log you can send to your SIEM.

Connects to
File shares, S3-compatible storage, PostgreSQL, SQL Server, REST APIs (OpenAPI), Microsoft 365, MCP servers

Sharq Engine

Trains a model on your data and serves it where your data lives.

Sharq Engine fine-tunes open models on your records and serves the result as a private endpoint. Your team can see every run, and reproduce it from its configuration.

Explore Sharq Engine
Open base modelYour model, trained on your records
Illustration. A fine-tuning run pulls a general model toward the way your organisation writes and decides.

  • Fine-tuning that fits

    LoRA adapters on open base models, so a specialist model costs a fraction of training from scratch.

  • One endpoint, training to serving

    The model you trained is the model you serve, through the same API, with lineage back to the run that made it.

  • Keys per person

    Every developer gets their own API key. Keys are stored only as hashes and can be revoked in the console.

  • Your GPUs or ours

    Runs on your hardware, or on Sharq capacity in-region.

Model families
Qwen, Llama, Mistral, Falcon, Jais

Start with one workflow.

A pilot runs on your own data, inside your boundary, and is measured against the numbers you already report.

Start a pilotRead the client briefing