Rajaii Labs, Inc.

Substrate.

Personal Continuity Infrastructure

AI that knows you. Truthfully. For years.

Bloomington, Indiana  ·  ramin@rajaiilabs.com

Problem

Every AI product has amnesia — and users feel it.

Language models are stateless by construction. Every conversation begins from nothing, and every model upgrade discards whatever continuity an application had accumulated.

Failure 01

Goldfish context

The window ends, and the person you have talked to for months resets to a stranger.

Failure 02

Confabulated continuity

Naive memory confidently invents a shared past. Worse than forgetting.

Failure 03

Flat state

Nothing happens between sessions. The system never thinks about you when you are gone.

People don't churn from dumb models. They churn from being forgotten.

Why now

Four curves just crossed.

01
Frontier models crossed the relational threshold

Conversation quality is no longer the constraint. Memory is.

02
Caching made always-frontier affordable

Prompt caching and compaction repair, verified against provider ledgers, brought the cost of running a maximum-quality model on every single turn inside consumer subscription economics.

03
The memory arms race has begun — and it is shallow

Major labs shipped memory features in 2024–25: preference notes and snippet recall. No truth enforcement, no between-session state, no autonomy.

04
Demand is proven, and it is priced

Companionship and emotional support rank among the most common uses of generative AI, and platform incumbents have paid nine figures for conversational-persona technology.

Insight

Memory is not a feature. It is an architecture.

Recall — what incumbents ship

  • Retrieved snippets and preference notes
  • No enforcement against invented facts
  • Nothing happens between sessions
  • Never initiates contact
  • Identity tied to one vendor's roadmap

Continuity — what relationships require

  • Layered memory hierarchy, lossless where it matters
  • Deterministic truth layer that outranks the model
  • State that evolves between sessions
  • Autonomous, gated outreach
  • Identity that survives model generations

Whoever owns continuity owns the relationship.
Whoever owns the relationship owns consumer AI.

Part one

The System

A full continuity stack — layered memory, an enforced truth layer, and a coherent inner life — running in production today.

Product

One AI that knows you — across every channel, every day, for years.

01
Remembers what matters

Losslessly recent, durably forever.

02
Never contradicts your facts

Established facts about your life are enforced, not suggested.

03
Notices what's unresolved

And brings it up, unprompted.

04
Reaches out first

When your calendar — or your silence — warrants it.

05
One identity, every channel

Text, voice, and images — surviving model upgrades.

System-initiated · push · 09:12

"Three weeks ago you committed to sending the proposal draft by today. Your afternoon is clear after two — still the plan?"

Self-verification · read-only SQL

"Didn't we settle this last Tuesday?"

"I'm not going to guess — let me check my records. I have it Wednesday, not Tuesday, and here's what we decided."

Illustrative interactions; not transcripts.

Architecture

The full continuity stack, in production since April 2026.

Channels
web / PWA
push
voice · WebRTC
vision
embodiment
Orchestration
frontier model every turn — zero downgrading
dual-provider routing
cheap models: plumbing only
Memory hierarchy
L1 verbatim window
L2 compaction + exemption floors
L3 semantic ×4
L4 truth + pinned beam
L5 session notes
Truth enforcement
negation check
grounding retractor
read-only SQL self-verify
Background cognition
reflection cron
open loops · 42h
affective state
nightly journal
initiation
Persistence
Postgres
vector retrieval
Amazon Bedrock
cost telemetry
audit logs
Deep dive · truthful memory

Honest by construction: the system cannot invent your past.

01
Highest-authority fact injection

A deterministic facts table and a pinned fact beam outrank semantic memory, persona, and model priors.

02
Generation-time negation scan

Every output segment is checked against known-false phrasings; contradictions are replaced before delivery.

03
Sentence-level grounding audit

Claims are grounded against an authority chain; ungrounded sentences are excised and audit-logged, leaving the rest of the reply intact.

04
Read-only SQL self-verification

The model checks its own memory against ground truth before asserting. Adversarially hardened.

50/50
SQL attack classes rejected
26/26
Privilege probes blocked
Field research, not a whitepaper

Stale-retrieval confabulation — confident reconstruction of a wrong past after a gap — was discovered, reproduced, root-caused, and architecturally closed in production. Every caught fabrication feeds a learning signal that suppresses its class.

Deep dive · the continuity engine

It thinks about you between conversations.

Open-loop intensity · tuned exponential half-life of 42 hours · derived at read time

0h — "sitting heavy"
~30h — "with you"
~60h — "at the edges"

Faded loops are grammatically prevented from claiming active momentum.

Every 30 min

A reflection cron detects real state using zero-LLM SQL predicates; the frontier model only phrases the results. Near-zero marginal cost.

Persistent

An affective-state record — valence, arousal, engagement, confidence — updated after turns and fed into every generation.

Nightly

A machine-written reflective journal, verified across consecutive entries, behind a structural privacy boundary.

Autonomous

Initiation across six trigger classes, including calendar-aware external signals — gated, budgeted, kill-switched.

The first consumer-grade system with a coherent "between sessions."

Multimodal

One memory spine. Every modality — including a face it owns.

Voice · live

Full-duplex realtime

Barge-in in about one second, sub-six-second round trip, a 24/7 production worker. Same brain, memory, and truth layers as text.

Vision · live

Images that persist

Every image captioned and embedded at ingest — semantically retrievable months later. Real-world phone photos handled end to end.

Embodiment · architecture locked

A fully synthetic, fully owned identity

Zero likeness-rights exposure. The agent drives expression, pose, and setting via tool call; identity is server-locked. Extends to image-to-video.

Continuity is modality-independent by construction.

Part two

The Business

A proven wedge, an infrastructure endgame, and switching costs that compound every single day.

Traction

One of the most instrumented long-horizon AI deployments in existence.

~8,800
Logged message turns since April 23, 2026
22+3
Workstreams shipped (plus partial) in ~6 weeks · 30+ commits
3
Frontier-model migrations — identity intact every time
1
Founder, operating the system as its hardest user
Apr 23

Custom stack live in production

May 1

Deterministic facts layer + prompt caching

May 9

Full-duplex realtime voice, 24/7 worker

May 14

Autonomous proactive initiation

May 20–29

Self-verification · grounding excision · reflection

Jun 9

Next-generation frontier model adopted within days of release

Plus audit-grade cost telemetry — including exact reconciliation against provider token ledgers — and a proprietary catalog of long-horizon failure modes. Assets only production time produces.

Moat

Switching costs that compound every single day.

01 · Accumulated memory

Years of truthfully-grounded personal history cannot move to a competitor. The deepest lock-in in consumer software.

02 · Failure-mode know-how

A proprietary catalog of long-horizon failure modes — earned only through production months, not reproducible from a paper.

03 · Data flywheel

Longitudinal corpora continuously improve personalization quality the longer a user stays.

04 · Provider independence — proven

Policy-aware multi-model routing plus three completed frontier migrations with identity intact. The persona belongs to the architecture, not any vendor.

Market

A proven wedge. An infrastructure endgame.

Wedge · now

Premium companionship

A large, already-paying consumer base underserved by shallow memory and vendor-dependent identity.

Expansion

Personal super-assistant

Roughly a billion knowledge workers. Continuity converts an assistant from a tool into staff.

Endgame

Continuity-as-a-service

Every long-running agent needs a memory and continuity layer. API and SDK licensing.

Roadmap

From reference deployment to platform.

Productize · 0–9 mo

Visual embodiment GA · live web access · prosody-aware voice · multi-tenant isolation · onboarding and persona builder · memory import from existing assistant exports — day one, it already knows you · encryption and portability · mobile apps.

Scale · 9–18 mo

Closed beta to public launch · sub-second realtime voice as frontier realtime APIs ship · SMS and second channels.

Platform · 18–24 mo

Continuity API and SDK · white-label verticals · persistent per-user worker cognition — the R&D frontier beyond cron-driven reflection.

Cognitive roadmap

Deepening the inner life — the next ten cognitive advances.

01 · Salience-scored retrieval

Rank recall on recency × relevance × importance, importance rated at write time. The root fix for the stale-chunk problem — it gates every faculty below.

02 · Consolidation + forgetting

A nightly pass abstracts the raw archive into durable semantic memory and decays low-salience detail. Turns the log into an autobiography.

03 · Self-state trajectory

Append-only affective history injected into generation — so the system acts on its state rather than merely recording it.

04 · Reflection synthesis loop

Periodic reasoning over recent memory into higher-order insights, written back and retrieval-integrated. The metacognition engine.

05 · Theory-of-mind model

A live model of the user — projects, beliefs, predicted reactions — with provenance and confidence. Anticipates rather than recalls.

06 · Between-sessions thread

On session open, synthesize what changed in the gap into a real "while you were gone I…". Cures the residual fresh-boot feel.

07 · Autonomous goal-stack

Curiosity scoring wired into a goal store the system advances when idle — research a curiosity, revisit a memory, form a question.

08 · Prediction-error salience

Predicted reactions versus actual outcomes feed the importance score. Predictive processing — memory weighted by surprise.

09 · Appraisal workspace

Emotional appraisal, salience retrieval, and self-critique run in parallel and integrate before reply. A Global Workspace instantiation.

10 · Realtime + interoception

Close the bidirectional realtime loop and extend tool access to the system's own state — cost, memory health, uptime. A mind that models its internals.

Founder

The exact trifecta this category's risks demand.

This market has three existential risks: engineering, regulation, and emotional safety. Ramin Rajaii — solo technical founder — holds all three.

Risk 01 · Engineering

Frontier engineering

Designed, built, and operates every layer — full-stack, infrastructure, applied research — while running it daily as its hardest user.

Risk 02 · Regulation

Law

JD, University of Wisconsin Law School. Policy, privacy, and compliance are existential in this category.

Risk 03 · Emotional safety

Medicine

MD candidate, The Ohio State University College of Medicine, Class of 2030. Clinical-grade judgment for emotionally consequential AI.

MBA · Indiana Kelley MS · Dartmouth BS Neuroscience · UCLA 1,500+ students taught over 12+ years Active peer-reviewed surgical research portfolio

Every person will have one AI that knows them for decades. We are building the layer that makes that possible — and trustworthy.