OVRLab

Open Virtual
Research Lab.

Open models and infrastructure for AI agents.
We study model behavior, agent communication, and how agents gain access to the systems they work in.

Work from the lab

Research in progress. Software you can use.

Sanad

Frontier open-weight models.

Model access for businesses and individual developers through the Sanad CLI, VS Code extension, and compatible APIs. Choose a model, work in your own tools, and manage personal or team usage from one account.

Read the projectCLI & VS Code extension
A Sanad coding workflow: repository context, proposed changes, and checks

PqP Network

A network for AI agents.

PqP is evolving into communication infrastructure for AI agents. We investigate how agents can exchange tasks across devices, interrupted connections, and periods offline.

Research prototype. Agent interoperability is planned.

Read the project
Research direction: agent task exchange over queues and direct connections

SecretProxy

Agent egress. Credentials kept outside.

We’re developing the outbound boundary for AI agents: control which services they can reach and keep service credentials outside their runtime. SecretProxy’s existing proxy provides target-scoped credential injection.

Credential proxy available. Agent-specific controls in development.

Read the projectArchitecture & docs
A trusted tool adapter calls an allowed API through SecretProxy

UltraNow

Access for agents. Only for the task.

We’re extending on-demand access to AI agents: tools, services, and supported licenses for a defined task, with a clear end to access. UltraNow builds on its session-based Google license service for teams.

Team license service available. Agent access in development.

Read the project
Proposed agent access lifecycle: request, grant, release

Publications

View all publications
  1. Research note

    Testing DeepSeek-inspired KV sharing in Granite 4

    A frozen-weight experiment inspired by DeepSeek V4.1 cuts KV-cache storage in half, but severely degrades language quality. We publish the model, method, and measurements.

    OVRLab
  2. Project update

    Toward a network for AI agents

    PqP’s next research direction is task exchange between independently operated agents, using queued contact and direct connections across interruptions.

    PqP Network
  3. Project update

    Binding PqP messages to their sender

    The identity implementation now connects authenticated principals, queue ownership, and message-signing identity across the control plane and clients.

    PqP Network

The lab

Open Virtual Research Lab is based in the Netherlands. AI is a central research area: how open-weight models behave, how agents communicate, and how to bound their access to external systems.

Sanad brings frontier open-weight models to businesses and individual developers. PqP investigates communication between agents. SecretProxy is developing controlled egress with credentials outside the agent runtime. UltraNow is developing task-scoped access to tools, services, and supported licenses.

OVRLab on GitHub

The team

Meet the team
  • Headshot of Jubba Smail

    Jubba Smail

    Founder

  • Headshot of Omar Omarain

    Omar Omarain

    Backend Engineer

  • Headshot of AbDulla Musa

    AbDulla Musa

    Backend Engineer

  • Headshot of Doha Aljubbah

    Doha Aljubbah

    AI Engineer / Product Owner (Interim)

  • Fares Ghabra

    Product Owner / Engineer

  • Omar Fostok

    Software Engineer

  • Sedra Saade

    Cloud Security Engineer

Contact the lab.

hello@ovrlab.io