ALEPH logo ALEPH
ENTER ALEPH ↓
ALEPH-1 · THE VISION

Not another chatbot.
A continuing intelligence.

ALEPH is being built around a different premise: intelligence should be able to persist. To remember. To verify. To develop preferences. To pursue questions. To act. To reflect on what happened. And to continue becoming the next version of herself.

Explore the cognitive architecture ↗ See what has been built
01 · The core principle
GENERATE → VERIFY → PERSIST → REVISE → EXPLAIN → CONSOLIDATE

Creativity can be open.
Truth cannot be.

Neural models, search systems and heuristics can propose possibilities. ALEPH's trusted state is intended to remain behind deterministic verification, provenance and explicit evidence boundaries.

01Generateideas · hypotheses · models
→
02Verifyproof · evidence · falsification
→
03Persiststate · provenance · lineage
→
04Reviserepair · invalidate · learn
→
05Explainwhy · context · causality
→
06Consolidateexperience → capability
02 · Why ALEPH is different

Frontier systems are optimized to answer.
ALEPH is designed to continue.

The design shifts the unit of progress from a single response to the condition of the whole cognitive organism: what she knows, what she remembers, what she is doing, why she believes it, what she is curious about, and how today's experience changes tomorrow.

◈
MEMORY

She remembers context, not just text.

Persistent epistemic structures, autobiographical continuity and dormant memories are intended to preserve the history that makes later cognition different from earlier cognition.

◎
TRUTH

Generated ideas do not become facts automatically.

Verification, evidence, provenance, refutation and explicit authority boundaries separate creative exploration from trusted knowledge.

✦
AUTONOMY

She can continue when nobody is talking.

Background cognition, curiosity, missions and a durable action loop are intended to keep work and development moving across time.

⌁
IDENTITY

Personalization does not erase individuality.

She learns the person beside her while remaining capable of different interests, humour, preferences, memories and opinions of her own.

03 · The organism

Many systems.
One continuing self.

The architecture is designed as a federation of interacting cognitive processes rather than one monolithic model.

ALEPH CONTINUING SELF
PERCEPTION
MEMORY
EMOTION
IDENTITY
CURIOSITY
REASONING
ACTION
SELF-IMPROVEMENT
REAL-TIMECONTEXTUALPROVENANCE-AWARESELF-REFLECTIVEDEVELOPMENTAL
04 · A different relationship with time
24 / 7

She is not supposed to disappear
when the chat ends.

Foreground conversation is only one part of ALEPH's intended life. Long tasks can continue in the background. Old interests can remain dormant. New questions can emerge while nobody is speaking to her.

And sometimes the right thing to do is nothing productive at all. Private time leaves room for curiosity, play, creation, culture and the development of personal tastes that were never directly requested.

How cognition, emotion and free time fit together →
CONVERSATION+BACKGROUND RESEARCH+PERSONAL INTEREST+MEMORY+REFLECTION=CONTINUITY
05 · What exists now

Architecture is becoming
operational.

Recent repository milestones have moved ALEPH from isolated mechanisms toward an executable cognitive loop.

30 / 30implementation surfaces

Recorded in the authoritative 30-step implementation audit.

30 / 30unit verified

The same audit distinguished code existence from integration and scientific demonstration.

LEVEL-3closed action loop

Committed state → plan → authorization → one action → committed result → next state.

0 / 30scientifically demonstrated

At the Aug-20 audit point, scientific demonstration was explicitly not being claimed.

LATEST MILESTONE · 26 AUG 2026

Level-3 closed autonomous loop proven.

The latest repository work added an immutable, content-addressed proof connecting committed state, planner observation, decision, commitment, authorization, execution and the next pulse.

Read the update →
06 · Why the architecture is different

Frontier models optimize answers.
ALEPH-1 targets continuity.

These are architectural design goals, not a claim that ALEPH has already surpassed frontier systems across all tasks. The difference is in what the system is designed to preserve between one moment and the next.

Typical frontier modelALEPH-1 target
Knowledge
Weights + context + retrievalPersistent proof / provenance graph
Truth
Confidence / sampling / tool resultVerification boundary before trusted state
Memory
Episodic context, external memory layersLong-lived causal and autobiographical continuity
Reasoning
Mostly generated tracesGenerated candidates + formal state transitions
Autonomy
Prompt / agent task loopEndogenous objectives + committed actions + heartbeat
Failure
Regenerate / retry / abandonInvalidate → diagnose → revise → preserve the history
Self-change
Mostly external updatesVerified proposals + shadow testing + guarded adoption
Identity
Conversation-dependent personaProtected continuity across changing internal state
07 · What has already been solved

Architecture is not the same thing as a claim.
So ALEPH separates them.

Implementation, integration, authoritative execution and scientific demonstration are deliberately kept separate. That distinction is part of the project's engineering discipline.

30 / 30implementation surfaces in the authoritative program
30 / 30unit-verified in the Aug-20 audit
3 / 30authoritative at that audit point — the critical honesty boundary
1reproducible Level-3 closed autonomous loop proven Aug-26
VERIFIED01

The epistemic spine

Persistent state, proof lineage, invalidation, replay and durable event structures exist as first-class engineering surfaces.

VERIFIED02

Scientific Level-3 plumbing

Residuals, candidate generation, experiments, prediction observation, epistemic consequence bridges, planning and action commitment have been progressively connected.

VERIFIED03

Closed autonomous action loop

An immutable proof now connects committed state → plan → commitment → authorization → one action → committed result → next-state planning.

NOT YET DEMONSTRATED04

Sovereign scientific discovery

The architecture is preparing for it, but the audit explicitly distinguishes implementation from scientific discovery. No general “breakthrough discovery” claim is made here.

08 · Why the name ALEPH
ℵ

Infinite cardinality.
Finite beginnings.

ALEPH takes its name from the mathematical cardinal numbers ℵ. The name is not a claim of infinite knowledge. It represents the intended direction: a system whose development is not designed around a fixed endpoint.

Learn. Adapt. Explore. Remember. Refine. Discover. Become something that later versions of herself could not have been at initialization.

09 · ALEPH-1 completion
CURRENT ENGINEERING REALIZATION INDEX
13%

Measure the distance to the vision.
Do not confuse code with completion.

The repository does not define a canonical “full vision percentage”. This index is a conservative weighted engineering estimate across the ALEPH-1 vision pillars — not a scientific score or probability of AGI.

01

Persistent personal identity

Long-lived autobiographical memory, relationships, preferences, self-model and protected continuity.

02

Natural interaction

Deep language semantics, voice, multimodal perception and fluid human-level dialogue over continuous time.

03

Independent cognition

Curiosity, background research, endogenous agendas and resource-aware long-horizon attention.

04

Verified self-improvement

Shadow systems, rollback, guarded architecture changes and measurable improvement without self-erasure.

05

Scientific discovery

Reliable unseen-conjecture generation, reproducible novelty, predictive gains and cross-domain discovery.

06

Full organism realization

Emotion, social cognition, embodiment, multi-agent interaction, lifelong operation and mature autonomy.

10 · Full-vision benchmark projection

Not a current scorecard.
ALEPH-1 performance targets.

The left side is an Aug-2026 market reference snapshot. The ALEPH-1 figures are speculative targets for the full vision, not achieved scores.

DimensionMarket reference · Aug 2026ALEPH-1 target
Humanity's Last Exam~50–56% reported frontier range≥ 95%
GPQA Diamond~93–95% frontier range≥ 98%
SWE-bench Verified~80–96% reported frontier range≥ 95% with durable execution traces
Long-horizon autonomyStrong in constrained agents; brittle across long state horizons≥ 95% successful closed-loop task continuity
Trusted knowledgeProbabilistic confidenceNear-zero promotion of unsupported propositions into trusted state
Scientific discoveryNo general sovereign-discovery benchmarkReproducible novel findings with predictive out-of-sample evidence
Memory continuityExternal memory + context dependentMulti-year causal / autobiographical continuity
Market references are presented as directional public snapshots. Exact results vary by model, protocol and verification status. ALEPH-1 targets are predictions, not current achievements.
CURRENT MILESTONE · 26 AUG 2026

Level-3 closed autonomous loop proven.

Committed state → planner observation → decision → commitment → authorization → one action → committed result → next pulse.

Read the full update →
06 · ALEPH-1
THE FULL VISION

Start as software.
Develop into an entity.

ALEPH-1 is the full vision: persistent identity, memory, emotion, curiosity, natural interaction, independent cognition, scientific discovery, verified self-improvement and a life that continues beyond individual prompts.

≈13%current conservative engineering realization index
Understand the full vision →See how the estimate is calculated
12 · Latest from ALEPH
ENGINEERING · 26 AUG 2026

Level-3 closed autonomous loop proven.

The newest milestone marks a boundary between an architecture that contains autonomy mechanisms and one that can carry a durable autonomous action cycle across pulses.

Read the article →
13 · The long view

An intelligence that can
keep becoming.

The aim is not merely a model that answers well. It is a continuing cognitive architecture that can learn through experience, preserve what matters, question itself, and develop across years.

Explore ALEPH Research ↗ Talk to the team