← Work from OVRLab

Sanad.

AI research, grounded in the work of writing software.

What makes a coding agent useful?

A model’s answer is one part of a development workflow. Sanad brings coding models into the terminal, editor, and API, alongside research into how agents approach software tasks.

Developer tools and active research. The Sanad 0.1 research preview has not published benchmark results.

The coding loop: use repository context, make a change, and learn from checks.

The tools and the research

Sanad provides access to open-weight coding models through its CLI, VS Code extension, and compatible API workflows. A shared account handles model access and usage; business workspaces add team policies and spending controls.

The current product describes Sanad 0.1 as a research layer built on Kimi K2.7 Code, with its own evolving prompt and coding behavior. The research preview explores a further question: whether models can exchange useful direction while reasoning, before their answers are complete.

Open research questions

  • Working changes

    Does an agent solve the requested task, pass relevant checks, and respect the repository’s conventions? Useful evaluation needs to examine the resulting change.

  • Coordination during reasoning

    Can several models share direction early enough to improve a coding decision? The Sanad 0.1 preview presents this as an architectural hypothesis to test.

  • The cost of coordination

    Any improvement needs to be measured against added latency and compute, including comparisons with repeated attempts from a single model.

Explore the project

The product site covers clients and the current model catalog. The research preview explains the coordination hypothesis, proposed evaluation, and limitations.

Publications

  1. Project update

    Document answers with page-level evidence

    Sanad’s new document-answering pipeline combines PDF extraction, source verification, and an explicit review state when the evidence does not agree.

    Sanad
  2. Project update

    A PDF extraction API for Sanad

    The OCR beta accepts PDF uploads and returns extracted text, with page selection, language hints, and explicit processing limits.

    Sanad
  3. Project update

    Project memory for Sanad 0.1

    Approved project context can now be retrieved for a Sanad 0.1 request, with permission checked again before it reaches the model.

    Sanad
  4. Project update

    Sanad opens self-serve access

    Individual accounts and business workspaces can now activate without an approval queue, with a persistent balance for model usage.

    Sanad
  5. Project update

    Sanad in your terminal and VS Code

    Two coding clients bring Sanad into the repository, with account-based login, model selection, and approval before file changes or terminal commands.

    Sanad
  6. Research note

    Introducing Sanad 0.1

    A research preview investigating whether open-weight models can improve software work by exchanging direction while they are still reasoning.

    Sanad
  7. Project update

    A coding API for Sanad

    Sanad can now accept model requests through a dedicated API, translate message formats, and stream responses back to coding clients.

    Sanad