No internet required.
The core companion keeps working when the network disappears.
Personal AI that lives with you, not above you.
Meet Coquí, a personal AI companion designed to stay with you, learn your rhythms, and act when you ask. No app to open. No distant server holding the relationship.
NATIVO / THE PROMISE
Nativo means native to the device. Intelligence belongs on the hardware, and to the person carrying it.
The core companion keeps working when the network disappears.
Personal context belongs on your device.
You own the relationship.
01 / COQUÍ
Clip it on. Keep moving. Coquí is designed for quick questions, useful nudges, private memories, and small actions between screens.

FROG SIGNAL / SECONDARY DIRECTION
The frog trace turns Coquí’s island namesake into a quiet visual signal.
COQUÍ · THE VOICE
Design explorations for the companion. Drag to rotate. Switch directions to compare.
Get oriented without opening a screen.
Useful when it should be. Fun when it can be.
Keep the meaning.
Actions happen with you in control of what comes next.
02 / THE SYSTEM
The foundation exists today. The adaptive scheduler is a research direction. The commercial intelligence layer remains private.
JibarOS is Android 16 with twelve AI capabilities exposed through one Binder AIDL interface. oird handles model lifecycle, queues, cancellation, memory budgets, and backend routing across llama.cpp, sherpa-onnx, ONNX Runtime, and libmtmd.
Governor is the proposed scheduler for concurrent models: balancing deadlines, memory bandwidth, power, and thermal headroom across CPU, GPU, and NPU resources.
AgentKit is the commercial intelligence layer for memory, tools, and orchestration: the private brain built on top of an open platform.
03 / WORKING PROOF
The physical companion is in development. The local inference foundation is already visible in code and in a working Android demonstration.
LLM completion, translation, Whisper transcription, and object detection, running concurrently in an Android Cuttlefish virtual device. Zero cloud calls.
Explore the current platform and its on-device capability architecture.
C++ / INFERENCE DAEMONShared model residency, priority-aware queues, KV-cache accounting. Load once, serve many.
ANDROID 15 / WORKING REFERENCESee local LLMs, MCP-aware apps, consent, and audit at the platform tier.
04 / ONE SYSTEM, THREE NAMES
The company, the platform, and the product form one deliberate architecture.
Native to the device.The company building personal AI that runs where it belongs: on the hardware itself.
Rooted. Resourceful. Self-reliant.Named for the jíbaro, the operating system makes AI part of the platform.
The keychain companion.Named for Puerto Rico’s small, unmistakable voice, Coquí is where the platform becomes personal.
05 / THE MARKET GAP
The nearest alternatives still depend on a remote service, a companion app, or both. Coquí keeps the complete companion loop on the hardware.
POSITIONING SNAPSHOT / SEPTEMBER 2026 · PRODUCT NAMES LINK TO SOURCES
06 / WHY NATIVO EXISTS
One company mission. One larger bet on where world-leading technology can come from, and where its value can stay.
Make AI native.Offline, private AI/ML that runs on-device. Coquí is the first proof of the mission.
Own the AI you carry.Portable edge AI, owned by the person who carries it.
Build it here.Prove the industry wrong from Puerto Rico. Build the company our engineers do not have to leave.
REVERSE THE FLOW
07 / FOUNDER–MARKET FIT
Rafael brings years across defense, public-safety, and surveillance technology: geographic information systems at CACI, public-safety radios at Motorola, and edge AI at scale at Flock Safety.
JibarOS and AAOSP are shipped proof that he can build on-device agent infrastructure. He built the systems that watch. Now he is building the one that does not.
That is why Nativo’s privacy is architecture, not marketing.
08 / THE SIGNAL STARTS HERE
The clearest view into the architecture behind Coquí is already public.
Explore JibarOSCompany one-pager