Magic Unicorn Special Projects researches, prototypes, and ships unconventional technology across AI, automation, infrastructure, hardware acceleration, and entirely new product categories.
Some become products. Some become open-source infrastructure. Some solve one especially stubborn problem. All of them begin with a question worth testing.
The shared foundation for distributed AI, agents, identity, memory, orchestration, deployment, and the specialized applications riding above it.
Private meeting intelligence with recording, transcription, speaker identification, searchable organizational memory, and action-item handoff.
Project operations designed for people and agents to plan, execute, document, and transfer work without losing the reasoning behind it.
A private health-intelligence workspace for exploring records, medical literature, timelines, and questions with clinician-aware workflows.
Our hardware platform for private, local AI — a compact NPU + iGPU box that runs a full 35B model on-device. Demonstrated as Home Ops, Meeting-Ops, and more: useful private agents without a datacenter or a surveillance business model.
We are most useful when the destination is important, the path is uncertain, and ordinary implementation recipes stop being useful.
Agentic workflows, retrieval, speech, vision, documents, knowledge graphs, evaluation, and private inference.
Product discovery, architecture, web applications, internal tools, SaaS, self-hosted deployments, and integrations.
GPU, NPU, iGPU, model serving, routing, observability, distributed inference, and odd hardware made useful again.
Technical reconnaissance, proofs of concept, stalled-project recovery, unusual integrations, and category invention.
Composable stacks, reproducible research, community-backed components, contribution plans, and sustainable licensing.
Technical due diligence, AI adoption planning, platform modernization, fractional leadership, and commercialization paths.
One platform underneath, a family of products on top. Unicorn Commander owns identity, memory, orchestration, model routing, and deployment — so every specialized app (Meeting-Ops, Project-Ops, and the rest) focuses on its own job while sharing one organizational brain.
Every engagement is different, but the discipline is consistent: reduce uncertainty, build evidence, and keep the useful parts.
Define the real problem, constraints, existing assets, prior attempts, and what success would mean.
Research the landscape, inspect the system, test assumptions, and identify the fastest honest path to evidence.
Build the smallest working system capable of proving or disproving the core hypothesis.
Put it in the actual environment where users, constraints, and entropy can express their opinions.
Deploy it, incubate it, open source it, license it, transfer it, or use what we learned to make a better decision.
Reproducibility, inspectability, and community participation make systems stronger. They also prevent the future from being rented back to everyone by the API call.
Build openly where it creates more capability, trust, and freedom.
Projects can mix open-source foundations, public research, commercial products, private deployments, and partner-owned work without pretending every problem has the same license.
Short technical reports can establish credibility faster than generic thought leadership, especially when they include configurations, measurements, and what changed.
UC-1 is our hardware platform for private, local AI — a compact NPU + iGPU machine that runs a full 35B-parameter model entirely on-device, no datacenter and no surveillance business model. On the same box we've stood up several demonstrated builds — Home Ops, Meeting-Ops, and more. We publish the model choices, the memory tradeoffs, and the throughput we actually measured.
Read the build log ↗Meetings, projects, contacts, customers, and agents all run over one connected organizational memory. That shared brain is what turns a pile of apps into a suite — and it quietly moves the product boundary.
Read the build log ↗Publish honest observations about VRAM, PCIe constraints, MoE models, batching, thermals, and workload placement.
Magic Unicorn is the company. Unicorn Commander is the platform. Special Projects is where uncertain ideas become working systems.
Not every project must run on Unicorn Commander, but the platform becomes the preferred foundation when shared agents, identity, memory, infrastructure, or deployment are useful.
Describe the mission, what has already been attempted, and what the first useful proof would look like. The form creates a structured email locally; no mysterious form backend required.
