EMPHOS Product Stack

Every product. One standard.

Local-first intelligence, premium execution, and a no-subscription model across every layer of the EMPHOS stack. From the flagship desktop assistant to the AI code editor to the knowledge system that thinks without being asked — each product is independent, and each one is stronger for being part of the same ecosystem.

Flagship Product
Available Now

Haven — Engineered for Presence.

The core intelligence layer of EMPHOS. A proactive desktop AI that operates locally, anticipates your workflow, and acts on your behalf — without a single server call. 23 capability domains, sub-2-second voice response, 100% private, and priced once.

23Capability Domains
<2sVoice Response
6Hardware Tiers
100%Local — No Cloud

The Living Sphere

Haven's visual presence — animated, responsive, aware. 10 avatar colours. Not a chat interface. Not a window. A presence that lives on your desktop and responds to your state.

Proactive Intelligence

Morning briefings. Pre-meeting prep. End-of-day summaries. Haven acts before you ask — because it knows your schedule, your files, and your patterns.

Browser Bridge + Plugins

Haven reads your active browser tab via the Haven Browser Bridge extension. Full plugin architecture with manifest-based capabilities. Plugin marketplace planned for Haven 0.2.

$79.99USD · One-time · Own it forever
New — Near Release

CAMS Code — The AI code editor that never calls home.

Local-first. AICL-powered. No subscription. No cloud. Your code never leaves your machine — and with AICL active, completions arrive 2,036ms faster than raw inference on the same hardware.

Qwen2.5-Coder Models

3B, 7B, 14B, and 32B parameter models in Q4_K_M and Q5_K_M. Six hardware tiers detected automatically at first launch. The right model for your machine, every time, no configuration.

AICL Protocol Stack

Sub-300μs classifier. Sub-500μs compiler. 5-lane routing across all conditions. Measured across 585 real inference runs: CAMS Code with AICL is 2,036ms faster than raw inference on average.

Offline · One-Time · Yours

Ed25519 offline license signing — verified locally, no server call required. Works on air-gapped machines. ~850MB installer. Buy it once. Own it permanently. Works if EMPHOS servers went dark tomorrow.

A direct local-first alternative to Cursor and GitHub Copilot — no monthly fee, no cloud inference, no code leaving your machine. Installer, signing infrastructure, and test suite complete. Near release.
The Protocol Layer

The infrastructure beneath the products.

Three patent-pending protocols power the EMPHOS stack. They are not products you buy — they are what makes every EMPHOS product smarter, faster, and more efficient than it would be without them.

Patent Pending
FTIP — Persistent State Channel
Eliminates 52–65 tokens per inference
Exactly 12 tokens per fractional sequence
KV cache as communication layer

Replaces repeated context injection with a persistent state channel — the LLM is a wire, not a brain.

Patent Pending
PSIP — Signal-Aware Contracts
Saves 47–61 tokens per response
1,680+ controlled inferences
10 topics · 5 signal types

Pre-negotiated signal contracts that compress what the model needs to say before it says it.

Patent Pending
AICL — Inference Router
Sub-300μs classifier · Sub-500μs compiler
585 runs · 11,389 tokens saved
5 test conditions · 4 models validated

Routes every query to the right protocol and tier in under 500 microseconds. Decoder in under 2ms.

Measured across 585 real inference runs: AICL active is 2,036 milliseconds faster on average than raw inference on the same hardware. That is not a latency reduction. That is a different experience.
Current & Near-Term Products

Built on Haven's core. Serving different people.

Active Development

PRISM

AI tutoring for children running locally on Llama 3.2 3B. Six learning modes. A hard-gate safety layer that logs every event — check phase, risk level, escalation flag — and cannot be overridden.

Per-skill mastery tracking with confidence scores and success rates. Emotion and attention detection on every interaction. Structured learning reports with highlights, concerns, and recommendations.

Active · Llama 3.2 3B local · Mastery tracking · Safety logging · Emotion detection
Models Trained & Exported

ETTS — Haven TTS Studio

Trained acoustic model and vocoder exported as fp16 ONNX — haven_am_fp16.onnx and haven_voc_fp16.onnx. Three deployment targets: Edge, GPU, TensorRT. VITS-based architecture. KL divergence verified byte-for-byte against the reference implementation.

Five Haven voice presets: Calm · Empathetic · Enthusiastic · Focused · Warm.

Trained ONNX models · fp16 GPU-ready · Edge / GPU / TensorRT · 5 presets
Patent Pending

VOXIS Lab

Zero-parameter voice synthesis derived from physics, not weights. The Acoustic Invariant Table (AIT) is VOXIS's Word2Vec for speech. Sub-500ms synthesis on CPU. 15,037-word corpus. No GPU required.

AICL urgency signal directly controls VOXIS speech speed — the protocol stack and voice layer are integrated at the physics level.

Sub-500ms CPU · Zero parameters · AICL integrated · Patent pending
Research & Intelligence

The long game. The moat that compounds.

Active Development · Step 14

HEINRICH Intelligence

A frequency-field knowledge system with a hybrid architecture — frequency-field semantic navigation at the core, a transformer Binding Layer (~2.1M parameters) for relational role assignment. It navigates semantic coordinates. It does not predict tokens.

507 tests passing. 55 confirmed universal semantic anchors — 31 at universality = 1.0, all 4 architectures agreeing with basin depths above 0.98. Step 14: Spontaneous Resonance — the field detects harmonic gaps and generates candidates for concepts physics implies should exist.

55Universal Anchors
507Tests Passing
527.7MB Knowledge Base
Active Research

EMPHOS Labs

The research engine behind the EMPHOS protocol stack. A 10-tab multi-protocol LLM inference platform with 10 active protocols: RAW, PSIP, FTIP, AICL, VIMP, WISP, PRIM, VMAP, APEX, DLP.

396 APEX genomes bred. Template bank of 1,822 queries across 6 categories. 30,000+ controlled observations recorded. Every protocol in the EMPHOS stack was proven here before it was shipped.

30K+Observations
542.6MB Research DB
10Active Protocols
Roadmap

What comes after Haven.

Every entry on this roadmap is funded by Haven's revenue — one purchase at a time.

2026
Now

Haven v1.0

Commercial launch. 23 domains, living sphere, AICL protocols, sub-2s voice, $79.99 one-time. The product that earns everything else.

Near
Release

CAMS Code

Local-first AI code editor. Qwen2.5-Coder 3B–32B. AICL-powered. 2,036ms faster than raw inference. No subscription. No cloud. Near release.

2027

PRISM

AI tutoring for children. Already running locally on Llama 3.2 3B. Mastery tracking, safety logging, emotion detection. First EMPHOS product designed for schools.

2028

Atlas

Mobile intelligence layer. Haven's proactive system extended beyond the desktop — real-time briefings, remote workflow control, continuity wherever you are.

2030

Shield

Security, privacy, and access control across the full EMPHOS ecosystem. Trustworthy by architecture, not by policy. Foundation for hardware integrations.

5–10 Yrs

EMPHOS HQ — Fraser Valley

Earthship architecture. Three levels. Solar powered. Thermally cooled by the hillside it sits in. Built from revenue. Built from the land. Built to last.

Everything starts with Haven.

The full stack begins with one product, already built, already running, already waiting. $79.99. One time. Yours forever.