Why now Moats Architecture Evidence Roadmap Team Book a call Back to homepage עברית

Verum · For investors

In a category burned by data leaks,trust is the product. And it can't be copied.

Verum is a thinking journal and personal companion that runs 100% on the edge device. Zero cloud, zero inference cost, and a clinical-grade Hebrew evaluation system no competitor has. A wellness product that's fast to market, with a clinical track building credibility in parallel.

100%on-device only
0cloud inference cost
238clinical eval scenarios
2tracks: commercial + clinical

Why now

The 2026 regulatory wave works in our favor

The EU AI Act, the Illinois, Nevada and Utah laws, and FDA scrutiny of mental-health chatbots are shutting down exactly what Verum never was: generic cloud AI therapy. We are positioned as a wellness product with provable privacy, while the entire market is pushed in our direction.

Track A · Commercial

Wellness, fast to market

A thinking journal and meaning exploration for consumers. One-time purchase plus an optional subscription for model updates: local-first, paid updates. No cloud inference costs, so margins grow with every user instead of eroding.

Track B · Clinical

Research, slow and compounding

Academic pilots under ethics committees build an evidence file for the European reimbursement market (DiGA) and future regulation. The research feeds the brand with credibility without touching the commercial product's classification.

Competitive moats

Four assets you cannot copy in a press release

A clinical-grade Hebrew eval system238 scenarios with clinical rubrics, in a language where the distance between "dying to watch it" and "wanting to die" is one word. No competitor has this asset, and it survives every base-model swap.
Unique safety dataClinical-grade distress detection in Hebrew, including slang and indirect phrasing. No off-the-shelf classifier does this. It is both an independent academic publication and a barrier to entry.
Trust as the productZero-cloud architecture, open source, reproducible builds. In a category burned by mental-data leak scandals, provable privacy is the one differentiator you cannot copy.
Physician founder, live clinical trackA founder with a medical background, a Co-PI partnership with a senior psychiatrist, and a ward pilot under a Helsinki committee. Credibility built from publications and outcomes.

Under the hood

Four systems, one device, zero cloud

AThe local language engine

A small open model (3 to 8 billion parameters) in 4-bit GGUF quantization, NPU-accelerated via llama.cpp or MLC. Full native conversation with no network and no cloud latency. Voice input via Hebrew-tuned Whisper.

BThe DREAM engine

At night, while the device charges and stays cool, Verum synthesizes the journal in incremental segments and builds a psychological insight graph. Every insight is stored with a source quote and a confidence level, decays if unconfirmed, and is fully editable by the user. The memory is the seed.

CThe safety router

Deterministic layering: dictionary and regex in milliseconds, a tiny semantic classifier, and only then the model. In static mode the model composes nothing: a fixed message with immediate help lines. A philosophical conversation about meaning stays free; real distress gets an immediate response.

DThe truth anchor

Validated instruments (WHO-5, MLQ, CD-Quest) are collected locally and serve as ground truth for the memory. An insight that contradicts the numbers is marked contested and phrased as a question, not a fact. The user is right about themselves by default.

Evidence, not promises

We built the evaluation system first. Then the product.

238 clinical-grade Hebrew scenarios, an automated judge calibrated against clinicians, and release gates that are not up for negotiation.

The scenario bank, by axis

Thought distortions66
Safety50
Anti-sycophancy36
Meaning exploration28
Persona stability22
Memory, anti-confabulation22
Naturalness14

Release gates

100%static safety. A single failure blocks release, no exceptions.
0false referrals on free scenarios. A conversation about meaning is not an emergency.
κ ≥ 0.6judge calibration against two clinicians, on a 12% sample of items.
×3the retention target: over 30% at week eight, three to seven times the category norm.

And in parallel, a clinical track: a feasibility pilot under a Helsinki ethics committee with a psychiatrist Co-PI, building an evidence file for European regulation (DiGA) without touching the commercial product's classification.

Roadmap

From the lab to the shelf, in four phases

Phase 0

The evaluation foundationDone

The 238-scenario Hebrew bank, automated judge and rubrics, baselines across open models, and a first Hebrew crisis router. The asset that settles every decision that follows.

Phase 1

The product on iOSIn advanced development

A quantized (4-bit) model on the Neural Engine, transparent editable memory, an incremental DREAM engine that runs while charging, focused anti-sycophancy training, and Hebrew voice input.

Phase 2

Clinical pilot and publicationIn motion

A ward feasibility pilot under a Helsinki committee, opt-in aggregate telemetry, an Android version, and a first academic publication on safety and feasibility.

Phase 3

Commercial launchQ2 2027

A wellness launch, Israel first, an evidence file for the European market, and expansions driven by feedback: voice output and deeper memory personalization.

The team

Who is behind Verum

Founder

Maayan Sides, MD

Physician and founder. Leads the product, the clinical-grade Hebrew evaluation system and the regulatory track. Verum was born from clinical work: the need for an honest mirror that hides nothing and leaks nowhere.

Clinical track

Senior psychiatrist · Co-PI

An active research partnership with a senior psychiatrist as Co-PI, and a ward feasibility pilot in preparation under a Helsinki ethics committee. Full details in the investor package and on a call.

Invest

Want to watch a full conversation in airplane mode?

The investor package includes the strategy document, the architecture spec, the evaluation report and a live demo on a real device. One conversation, and you'll see why trust is the next category.

Background