Skip to content

Outcome Quality Engine

It learns how you think.
Then tells every AI how to answer you.

NOMARK reads your history, figures out how you actually reason, and writes the instructions your AI follows — so you stop re-explaining yourself every conversation.

See what it would say about youRead the Docs

The Shift

Memory tells your AI what you said. NOMARK tells it how to answer.

Every AI product is racing to remember more. That's the wrong problem. The problem isn't that your AI forgets what you said — it's that it doesn't know how you want to be answered. That's not a fact in your history. It's the shape of it.

Memory approach

Remembers what you said:

  • User asked about Stripe integration on March 12.
  • User works in TypeScript.
  • User dislikes long preambles.
  • User said "just give me the code" last week.

Your AI still has to guess what to do with any of it.

NOMARK

Writes how to answer:

  • Skip preamble; lead with the working code.
  • Assume TypeScript; do not re-explain type system.
  • Cite sources when making empirical claims.
  • Provide working draft over detailed plan.

Instructions the model follows, before you type a word.

The Discipline

It knows which ‘you’ is the real you.

A bad-mood opinion shouldn't rewrite your profile. A real shift should. NOMARK learns continuously, but tells one-offs apart from real patterns — and promotes signals only after they hold across three sessions or more.

Promotion gate — observed signals

  • Prefers working drafts over detailed plans.

    Observed across 14 sessions, 3 providers.

    Stable

  • Challenge assumptions proactively.

    Observed across 11 sessions.

    Stable

  • Wants more hand-holding today.

    One session. Mood shift or new context — not yet promoted.

    Staged

  • Prefers long narrative explanations.

    Contradicts 23 observations of the opposite. Not promoted.

    Rejected

Contradictions don't get averaged. They get resolved — by which signal held longer, across which providers, under which context. Old preferences aren't deleted when you change; they're dated.

The Trust

Every rule has a receipt.

Click any rule NOMARK writes about you and see the exact messages that taught it — which conversation, which provider, which date. Every instruction is backed by real evidence. If a rule is wrong, reject it and it's gone.

Rule

Prefer sourced claims over synthesis.

evidence_orientation · high · 34 observations · stable

Evidence — quoted from your own history

  • 2026-03-21

    “where does that number come from”

    claude
  • 2026-03-28

    “show me the actual data, not a summary”

    claude
  • 2026-04-02

    “is that from the paper or are you making it up”

    chatgpt
  • 2026-04-09

    “cite the source”

    claude

Nothing is inferred without evidence. You can reject any rule and it's gone.

Reject ruleKeep

The Instructions

Real instructions. Not labels.

Most personalization systems tag you evidence-oriented, concrete-first, convergent. Your AI can't do anything with that. NOMARK compiles tags into commands the model actually follows — before you type a word.

A real profile — 717 conversations

14 rules · 10 dimensions

I. Format

  • 01

    Default to terse, outcome-first responses.

    format_preference · structured · 34 obs

  • 02

    No preamble, no hedging language.

    format_preference · structured · 28 obs

  • 03

    Use structured artifacts over narrative when content is multifaceted.

    format_preference · structured · 22 obs

  • 04

    Cite sources when making empirical claims.

    evidence_orientation · high · 31 obs

II. Reasoning

  • 05

    Prefer sourced claims over synthesis.

    evidence_orientation · high · 34 obs

  • 06

    Surface assumptions explicitly.

    evidence_orientation · high · 34 obs

  • 07

    Allow provisional answers; do not force premature resolution.

    ambiguity_handling · tolerant · 28 obs

  • 08

    Lead with principles and frameworks; examples validate, not discover.

    abstraction_preference · abstract-first · 15 obs

  • 09

    Present trade-offs before recommendation; compress to 2–3 options.

    explore_vs_decide · convergent · 22 obs

  • 10

    Challenge assumptions proactively; offer disconfirming evidence.

    belief_revision · fluid · 8 obs

  • 11

    Surface downstream consequences and feedback loops unprompted.

    systems_framing · systemic · 12 obs

  • 12

    Derive from first principles; omit social proof.

    reasoning_autonomy · high · 31 obs

  • 13

    Compress to conclusion; minimize process narration.

    metacognitive_explicitness · low · 19 obs

  • 14

    Provide working draft over detailed plan.

    action_cadence · rapid-iterate · 14 obs

Generated from one person's history. Every rule traces back to exact quotes you can read — and reject. Sent to every model, every message.

The Evolution

Your thinking changes. So does your profile — without losing where you started.

Old preferences aren't deleted when you change; they're dated. You can see exactly how your thinking has evolved, dimension by dimension, over months. Every night, NOMARK cleans itself up — merges duplicates, retires contradictions, prunes noise.

Dimension timeline

How you sit with uncertainty

ambiguity_handling

  1. 2025-11

    0.22

    Initial read: wanted single answers, resolved fast.

  2. 2025-12

    0.28

    Pattern held across 8 sessions.

  3. 2026-01

    0.51

    Shift detected — "it depends" responses accepted more often. Staged.

  4. 2026-02

    0.56

    Promoted to stable after holding 3 sessions.

  5. 2026-03

    0.74

    Further shift — provisional answers explicitly requested. Re-promoted.

  6. 2026-04

    0.78

    Current — rules updated across all providers.

Six months. One dimension. Nothing overwritten, nothing lost. The instructions your AI receives today are cleaner than yesterday's — and you didn't touch anything.

The Panel

Your instrument panel.

See what AI knows about you. Prose reflection, preference bars, conflict detection, and governance — in one view.

Identity Mirror

You write like someone who has said the same thing too many times to too many models. Direct. Short. No preamble. When you say "make it shorter," you mean cut 40% and remove every hedge. Bullet points over paragraphs. Code over explanation. You have 15 correction patterns that repeat — the system has learned them all.

Tone: Direct
0.87
Length: Short
0.74
Format: Bullets
0.65
Jargon: Plain English
0.71
1 preference conflict detected

Cross-Platform

Your correction in Claude improves your Gemini output.

One ledger. 40 entries. Under 3K tokens. Every model you use reads the same preference profile — and gets better from every correction, everywhere.

C

Claude

847 conversations

G

ChatGPT

1,247 conversations

Ge

Gemini

312 conversations

Cu

Cursor

523 conversations

N

NOMARK Ledger

2,929

Conversations analyzed

147

Signals extracted

40

Ledger entries

2.4 KB

Total size

The Portability

Same you, every model.

Claude, ChatGPT, Gemini, local models — all of them receive the same compiled instructions. Switching providers stops meaning “start over.” Your reasoning style travels. Pick the entry point that fits your stack.

import { createResolver, parseLedger }
  from '@nomark-ai/engine'

const entries = parseLedger(ledgerContent)
const resolver = createResolver({ entries })
const result = resolver.resolveAll()
Apache 2.0·Open source·No account required·Works with Claude, GPT, Gemini, Llama, or local

Open Core

Free to start. Pro when you need it.

The engine is open source. The layers that enterprises need — trust contracts, team baselines, audit trails — are Pro.

Apache 2.0

Engine

Full preference resolution. Local-first. No account needed.

  • ✓Ledger parsing & resolution
  • ✓Meaning map extraction
  • ✓Signal decay & weighting
  • ✓Preference scoring
  • ✓CLI tools (import, profile, init)
  • ✓MCP server
  • ✓Python & TypeScript SDKs
View on GitHub
BSL 1.1

Pro

Trust, governance, and team intelligence. The layers enterprises need.

  • ✓Everything in Engine, plus:
  • ✓Weighted scorer with instability detection
  • ✓Trust contract enforcement
  • ✓Critical-field gates (advisory → hard block)
  • ✓Contradiction resolution engine
  • ✓Archive sync & cross-device ledger
  • ✓Team baselines & scoped overrides
  • ✓Audit trail & compliance reporting
See Pricing

Pricing

Simple, transparent pricing.

Start free. Scale when you need to.

Local

Freeforever

Run locally. No account.

  • ✓Ledger parsing
  • ✓CLI tools
  • ✓40 entries max
  • ✓Single device
Install

Free

$0/month

Cloud sync. One user.

  • ✓Everything in Local
  • ✓Cloud sync
  • ✓Web panel
  • ✓40 entries
  • ✓Email support
Sign Up

Most popular

Pro

$15/month

Full resolution engine.

  • ✓Unlimited entries
  • ✓Trust contracts
  • ✓Contradiction detection
  • ✓Priority support
  • ✓500 SU/month
Get Started

Scale

$85/month

High-volume individual use.

  • ✓Everything in Pro
  • ✓Batch resolution
  • ✓Instability detection
  • ✓Full audit trail
  • ✓2,000 SU/month
Get Started

Pipeline

$150/month

CI/CD and automation.

  • ✓Everything in Scale
  • ✓API access
  • ✓Pipeline gates
  • ✓Webhook imports
  • ✓10,000 SU/month
Get Started

Team

$20/user/month

Shared baselines. Team intelligence.

  • ✓Everything in Pro
  • ✓Team baselines
  • ✓Scoped overrides
  • ✓Admin dashboard
  • ✓2,000 SU/user/month
Get Started

Enterprise

Custom

Compliance. Audit. SLA.

  • ✓Everything in Team
  • ✓SSO/SAML
  • ✓Audit trail
  • ✓Data residency
  • ✓SLA
  • ✓Unlimited SU
Contact Sales
See full pricing details →

Your AI should already know this about you.

Stop re-explaining. Stop re-prompting. Build your preference profile once — every model you use reads it.

Get Started — Free
Open source·Local-first·Model-agnostic·Patent pending