PIXELSAFFRON

Applied AI researchEngineeringProducts

Building practical systems for artificial intelligence.

Pixelsaffron researches and builds open, practical AI technologies that help people create, control, test, and deploy intelligent systems.

Explore our workXorafin
19.0760°N · 72.8776°EPIXELSAFFRON // v0.1
01About

We work on the gap between AI capability and dependable real-world systems.

Artificial intelligence is becoming increasingly capable, but models alone do not create reliable products. Real-world AI also requires memory, tools, permissions, evaluation, observability, deployment, and human control.

Pixelsaffron researches these layers and turns the strongest ideas into practical products and open technologies. We are beginning with developer software and intend to expand gradually into deeper AI systems, specialised models, edge intelligence, and physical AI.

02Current work

Xorafin

live

Portable AI agents, treated like software.

Xorafin is a local-first platform for creating, running, testing, versioning, and sharing reusable AI agents. It helps developers move beyond one-off prompts and fragile scripts by packaging an agent's instructions, tools, memory, behaviour, and tests into a portable Agentfile.

CREATE
Define reusable agents
RUN
Execute locally in memory
INSPECT
Understand tools, memory, and traces
TEST
Measure reliability before deployment
SHARE
Version and distribute agents
Visit XorafinView the direction
research-agent · agentfile
name: research-agent
model: provider/model
tools:
  - web-search
  - document-reader
memory:
  type: local
tests:
  - research-quality
  - citation-accuracy
03Principles

How we choose what to build

  • 01

    Real problems before impressive demos

    We build around repeated, meaningful problems rather than temporary AI trends.

  • 02

    Control without unnecessary lock-in

    Users should retain control over their data, models, definitions, and deployment environments.

  • 03

    Open foundations, sustainable products

    Core standards and adoption layers can remain open while managed infrastructure, collaboration, governance, and scale support the business.

  • 04

    Reliability over magic

    AI systems should be inspectable, testable, observable, and honest about their limitations.

04Direction

Starting with agents. Building toward broader AI systems.

  1. NOW

    Reusable AI agents

    Xorafin, Agentfile, local execution, testing

  2. NEXT

    Agent infrastructure

    Collaboration, deployment, evaluation, governance

  3. LATER

    Specialised and open models

    Model adaptation, tool-use models, local intelligence

  4. LONG TERM

    Edge and physical AI

    Devices, distributed systems, robotics orchestration

This is a direction, not a promise to build every layer immediately. Each stage must be earned through real usage, technical progress, and a sustainable business.

05Open

Open where openness creates trust and progress.

Pixelsaffron intends to contribute open specifications, software, tools, research, evaluations, and selected models where doing so improves ownership, interoperability, and shared technical progress.

Commercial products will support the infrastructure, security, collaboration, governance, and operational responsibility required to maintain that work sustainably.

  • Open foundations.
  • Sustainable infrastructure.
  • No unnecessary lock-in.

We are beginning with one product and one meaningful problem. The ambition is larger, but the path is sequential.

Build something useful. Earn trust. Develop original technology. Then pursue harder problems.

Follow the work
  • USEFUL
  • DEPENDABLE
  • CONTROLLABLE
  • REUSABLE
  • OPEN
  • INTEROPERABLE
  • DEPLOYABLE