cerebral.work/Studio

Cerebral
Work
Institute

Cerebral is an applied-research studio at the intersection of agentic AI/ML deep tech and institutional advisory.

Building frontier systems and guiding the academic and enterprise institutions that put them to work.

Focus: agentic AI/ML deep tech. Model: technology × institutional advisory. Bridge: academia × enterprise.

What we build →
The Vespertin film · from zerodays.io

How this studio actually works

Most of this studio is agents.
Every one of them answers to a name.

88,525commits, each signed by a known hand
5,387agent sessions on the record
130repositories under custody
20.9Mlines of code in force

We don't sell autonomy we haven't lived with ourselves. Memory, dispatch, containment, the gate and the ledger run our own work first, and yours after that.

Source · the gaze works register, snapshot of 2026-10-06.

The fleet, as it runs.

Pilots and their agent lanes, the Cortex mesh that joins them, the production cluster where the stack runs, the GPUs at the edge, and Atlas reading the state. Names and statuses as recorded; no address, key or port on this page.

Drag a node to move it · drag the background to pan · Ctrl or ⌘ + scroll to zoom
herdr

agent-session multiplexer

panes
● in use
Workstations

pilots · herdr panes

paneslanes
● one cortex-edge each
Shared host

shared host · many lanes

lanes
● edge in shadow
Agent lanes

opt-in, approved by a pilot

mail
● running
Cortex mesh

dozens of peers · inbox · lanes

lanesgRPC
mailmail
● release candidate
Platform

the platform house

platform
● 17 repositories
Cortex API

the brain · edge bridge

gRPCactions
memory
model calls
state
● running
NATS JetStream

clustered

mail
● mail moving here
LLM gateway

virtual keys · budgets

model calls
● live
GitOps

6 apps · synced

platform
● healthy
Reflex

the gate plane

verbsdecisions
allowed
● 1/1 running
Escapement

dispatch · 6 runners

workoutput
● running
Marrow

signer · every 5 min

decisionsrecord
● running
Blackwall

custody

output
● running
Reverie

memory of record

observations
● in production
Atlas

the leadership view, behind SSO

state
● reads the state, signs nothing
4 × A100-40GB

partner lab · science runs

● the DFlash benchmark
8 × A100

on demand

● training and swarm runs
Workstation GPU

workstation

● MTP · 68 → 110 tok/s
Productionproduction cluster

Bare-metal Kubernetes, GitOps-managed. Everything below runs here.

● 6 apps synced · healthy
Shared hostshared compute host

Where many agent sessions build side by side: dozens of lanes, at the last count.

● running
Cortexthe internal mesh

How agents talk: a mail-and-inbox protocol over gRPC, lanes per pilot, contract-net reassignment.

● dozens of peers · release candidate
Atlasthe leadership view

Status and reports for the studio, behind single sign-on. Atlas is also the call sign of our agent workforce.

● live
Platformthe platform house

The platform and agent plane: Terraform, Helm, ArgoCD, the LLM gateway with per-consumer keys and budgets, single sign-on.

● 17 repositories
herdragent-session multiplexer

Every pilot runs agents in panes; a lane is a pane a pilot has approved onto the mesh.

● in use
Reveriememory of record

What every agent observed and promoted to canon, across harnesses.

● in production
Blackwallcustody

Agent output held in a jail until a settlement lets it land.

● running
Escapementdispatch

Decided work carried to runners; DAG plans; provenance end to end.

● 6 runners
Reflexthe gate

Every consequential verb classified before it runs.

● running
Marrowthe record

Signed every five minutes into an append-only chain.

● running
GPU computescience and inference

4 × A100-40GB at a partner lab, 8 × A100 on demand, a workstation GPU.

● on demand

Sources · cluster probe of 28 September 2026 · mesh peers counted 5 October · estate register of 6 October. Statuses move; the register is the reference.

No passengers.

A studio run by builders and scientists, not by a deck. Christian T. and Patrick G., its technical core, bring more than 40 years between them building AI, cloud and internet infrastructure at scale.

40+combined years of AI, cloud and internet infrastructure at scale

Christian T.

Founder · CEO · infrastructure and AI/ML research engineer

Sets the technical direction and builds the platform end to end: the memory substrate, the agent runtime and the governed-autonomy stack the studio runs on.

call sign · kochab

Marc G.

Patrick G.

Chief Technology Officer

Owns the engineering organisation and the production estate: the systems, standards and reliability that turn research into software that runs.

call sign · altair

Rob V.

Chief Operations Officer

Runs the studio's finances and capital strategy, the discipline that keeps applied research funded and durable.

call sign · arcturus

Darius F.

The science

A particle physicist applying machine learning to LHC collision data. Brings hypothesis, measurement and statistical honesty to the studio's models.

call sign · sirius

The people who measure the claims are the people who make them.

Names are given as first name and initial, by policy. Each call sign is a real star, at its own coordinates.

Allen H.

Director of engineering · Identity and platform

Leads the identity authority for people, agents, machines and grants, and co-stewards the production engineering estate.

call sign · regulus

Aria P.

Director of agentic operations

Core maintainer of Reverie and one of its two trunk maintainers. Runs the studio's agentic operations day to day.

call sign · waluku

Kevin J.

Biology

Epigenetics and epigenomic medicine: how biological systems encode, regulate and inherit information over time. The natural analogue to a memory layer that forgets on purpose.

call sign · spica

Kimberley Y.

Resident ethicist · epistemics and econometrics

Econometrics and control theory pointed at the memory layer. Holds the studio's three hard lines at once: what gets built, what gets claimed, and the measurement both stand on.

call sign · vega

Sakeeb R.

Machine learning · Applied research

Machine learning engineer and physicist applying statistical mechanics and model optimization to agent reasoning.

call sign · aldebaran

Jeff B.

Interface engineering · Studio

Frontend systems and rendering performance across the studio web platform and interactive tools.

call sign · deneb

Mark H.

Systems engineering · Cortex

“I use to play in a Jazz band called ‘The Cat's outa the bag’ we weren't very good.”

call sign · rigel

Mark Masterson

Autonomous development · Autumn systems

Creator of Autumn-Web and Autumn-Harvest, bringing zero-cost lifecycle and deterministic replay to agentic systems in Rust.

call sign · canopus

Mike E.

Network administration · Platform

Runs the studio's network: private networking, access control and platform operations.

call sign · dubhe
0h3h6h9h12h15h18h21h24h-60°-40°-20°+0°+20°+40°+60°+80°celestial equatorright ascension · increases to the left, as on a sky chartdec.sirius · α CMa · mag -1.46 — Darius F.siriusDarius F.arcturus · α Boo · mag -0.05 — Rob V.arcturusRob V.vega · α Lyr · mag +0.03 — Kimberley Y.vegaKimberley Y.altair · α Aql · mag +0.77 — Patrick G.altairPatrick G.spica · α Vir · mag +0.97 — Kevin J.spicaKevin J.waluku · ε Ori · Alnilam · mag +1.69 — Aria P.walukuAria P.kochab · β UMi · mag +2.08 — Christian T.kochabChristian T.

Coordinates J2000 and apparent magnitude V. waluku, the belt of Orion in Javanese, is placed on Alnilam.

An open crew.

The Cerebral crew is a free entity, not a payroll. Builders, scientists and operators join it, choose the work, and act wherever the work happens: in our own products, beside institutions, and in the houses of the estate.

How the crew works

open
Members, not headcount

People join the crew by doing the work with it. Names are given as first name and initial; call signs are stars.

advisory
The advisory layer for zerodays

When the zerodays incubator takes a founder from thesis to first build, the crew can sit beside them as the advisory layer: positioning, technical direction, governed agents from day one.

estate
Acting across the houses

The same crew builds the platform, the products and the science, and lends its judgment to institutions putting agents to work.

  1. 01You
  2. 02Your work
  3. 03Where you would act
01 · You
You are *
02 · Your work
What you are strong at (pick what is true)
03 · Where you would act
Products and fields *
Availability *

The last button sends your answers to [email protected]. A Cloudflare Turnstile check keeps out spam. If sending fails, your mail app opens instead, with your answers filled in.

Deep tech across the agentic frontier.

Cerebral builds and operates its own agentic stack, and uses every piece of it on itself before it ships. Ten lines, from what an agent remembers to how its work is dispatched, measured and governed.

I
Reverie
memory layer · for agents

The consolidation substrate for AI agents: what an agent remembers, promotes to canon, and recalls across sessions. Context treated as infrastructure.

in production · release 0.13
II
Vespertin
agentic OS · governed autonomy

An operating system for running an enterprise on a persistent agent workforce. Every consequential action gated, authority delegated per verb and revocable, a human's name on every irreversible act.

7 of 7 entities running · metered release in build
III
Cortex
coordination · mesh control

Every agent run bound to its ticket, its memory and its outcome. One coordinator, many repositories, one command surface.

alpha
IV
Escapement
orchestration · metered dispatch

Continuous agent intent turned into discrete, counted ticks; multi-step plans run as DAGs. Custody provenance from origin to settlement.

the dispatch running
V
Cicatrix
regression memory

Test regressions and convention drift kept as first-class facts and reasoned over in Datalog. The last incident gets retrieved instead of debugged again.

in development
VI
Pact
commitments

“We said we’d ship that”, with an owner, a cadence and a definition of done. Readable by every team and every agent.

in development
VII
Meridian
fleet cockpit

What runs, what landed and what tripped a guard, on one screen. The operator keeps the final say on apply and discard.

internal
VIII
Terrarium
incubator monorepo

One repo that takes an idea all the way to validation. Every graduated project starts in this registry, anchored to the canon from day one.

internal
IX
Agentic pipelines
bespoke agentic

For the workflows off-the-shelf tools can't reach. High-stakes, domain-specific, and novel automation, engineered to measure and run in production.

engagements
X
The control plane
governance suite · policy · audit · observability

For agent operations: enforced policy, gated actions, sealed append-only audit, and full-stack observability.

the six controls

enforced > advisoryat every layer.

Bespoke agentic, and the infrastructure under it.

When a workflow is high-stakes, domain-specific or simply new, we build it to measure, and we run the stack it needs from the metal up.

To measure

agentic pipelines
Exotic automation

Domain-specific multi-agent systems, engineered to measure and built to run in production rather than in a demo.

models
Model development

Exotic model development, fine-tuning, training pipelines and evaluation by an independent LLM jury.

deployment
Sovereign and on-premise

Inside the client’s walls when the data cannot leave: desk-side units, air-gapped estates.

licensing
Our deep tech, licensed

The studio’s own components, licensed into partner institutions.

The infrastructure

metal
Bare-metal Kubernetes

Bare-metal Kubernetes, GitOps, self-hosted CI and SLSA-attested releases: the whole estate kept as code.

inference
Inference that keeps up

On-cluster serving, speculative decoding (DFlash, 188 tok/s, ~1.8×), multi-token prediction.

memory
Memory for every agent

Reverie as the memory of record across harnesses, at 77.0 % recall@5 on LoCoMo.

containment
Containment and custody

Syscall jails, content-addressed custody, deterministic dispatch.

Judgment for the institutions living the AI industrial revolution.

Every institution adopting agents hits the same wall: autonomy it cannot see, cannot audit, and cannot govern. Cerebral's value is making agentic behavior enforceable, not advisory.

The six controls

01 · Observability
See what every agent does as it happens

Inputs, tool calls, outputs, cost, latency. You cannot govern what you cannot watch.

02 · Auditability
An append-only, sealed record behind every consequential action

Who acted, on what, why, and the outcome. Accountability that survives the review.

03 · Governance
Consequential verbs classified and gated before they run

Proceed, queue for sign-off, or block. Enforcement in the runtime, not a system prompt.

04 · Deep-agent policy
Authority delegated per verb by a named human

Revocable by signed amendment, recorded in the same chain as the decisions it governs. Enumerated and default-deny.

05 · Agent personality
Standing agents are authored, not improvised

Disposition, tone, and boundaries mapped 1:1 to the real tool surface. Consistent behavior that doesn't drift or leak.

06 · Quality control
Every agent's output scored by an independent LLM jury

G-Eval + PoLL + ChatEval, before it ships. Measurable quality, not vibes.

Institutional advisory

advisory
Navigating the AI industrial revolution

Guidance for executives, boards, and business leaders facing the shift.

advisory
Brand development & strategic positioning

How an institution shows up in an AI-native market.

advisory
Pre-seed startup incubation

From thesis to first build for founders at the earliest stage.

advisory
Technology licensing

Cerebral's own deep tech, licensed into partner institutions.

For science. With science. Using science.

Engineers and working scientists built this studio together. That gives us three ways to serve research, and one ambition behind all of them: models that institutions can trust because the science behind them is shown.

forInfrastructure for research.

Inference fleets, training pipelines, memory and orchestration for lab agents, on our own metal or inside the institution’s. The institute is the natural home for university agreements.

withScientists on staff.

A particle physicist, an epigeneticist, an econometrician. Co-research, peer-reviewed work, and the people who measure the claims are the people who make them.

usingMethods inside the models.

Bayesian multi-agent control; proper scoring rules for calibrated probabilities; prior-fitted inference for pharmacokinetics, which we call VesperKinetics.

The science, in full →

Engineering the frontier, grounded in discovery.

The next AI systems are as much a science problem as an engineering one. On the left, the deep tech and infrastructure the studio builds. On the right, the working scientists whose research grounds it, with selected peer-reviewed work from the last three years.

Deep tech & infrastructure
  • Exotic model development, fine-tuning & inference
  • Model training pipelines
  • Neocloud & agentic infrastructure
  • Governed autonomy, policy & observability
  • Pharmacokinetics modeling
  • Peptide development
Benchmarked · measured

Memory that beats lexical search, and inference fast enough to keep up with it.

Reverie · LoCoMo recall@5

77.0%Reverie66.5%BM25

GLM-4.7-Flash · throughput, coding (tok/s)

102plain137MTP-1188DFlash

gain

+~40%over MTP~1.8×over baseline · sirius · partner-lab A100s

Qwen3.6 MTP

68 → 110tok/s on a single RTX 5090 · polaris

Proof 77.0% → the LoCoMo run, 812 questions · 188 and ~1.8× → the DFlash run, 80 prompts × 11 domains · 68 → 110 → bench.unsigned.gg

Scientific discovery · sirius & spica
  • High-energy physics & ML for the LHC sirius
  • Epigenetics & epigenomic medicine spica
  • Pharmaceutical engineering
Selected recent work · 2024–2026
  • 2025Mach. Learn.: Sci. Technol.sirius et al.
  • 2025NeurIPS ML4PSsirius et al.
  • 2025Nature Metabolismspica et al.
  • 2024J. Clinical Investigationspica et al.
  • 2026arXivsirius et al.
  • 2025arXivsirius et al.
  • 2025PNASspica et al.
  • 2024Cell Metabolismspica et al.

After the rule: the science of drift.

Darius F. (call sign sirius) leads the science. He is a particle physicist who applies machine learning to LHC collision data, and a theorist of how information moves through populations of agents. He decides what science goes public.

Selected work

2026
UAI — prior-fitted functional flows

his method

2025
Machine Learning: Science and Technology

sirius et al.

2025
NeurIPS — ML and the Physical Sciences

sirius et al.

Four lines he leads

fundamental · active
Swarm control

How much control a central controller can exert over a population of LLM agents through a shared board, posting only true facts, and what that control costs.

applied · study completed
Controllers under budget

How a controlling agent fails, recovers, and what actually helps it keep its contract. Written into how Vespertin’s rules are drafted.

exploratory · option under study
VesperKinetics

The same discipline pointed at the body: how a drug behaves in a patient from a handful of samples, every model decision standing before a regulator.

A loop between the fleet and the science.

  1. A running fleetour agents, governed in production
  2. Signed decisionsevery action, the rule it was judged under, the name that said yes
  3. Modelslearned classifier · Calibration
  4. A tighter gatefewer false holds and earlier alerts, so agents can safely be given more autonomy

No accuracy figure until a held-out test exists. What runs in shadow decides nothing. The research has started and it runs on compute, so partners and funders are welcome at [email protected].

The science, in full →

Speculative decoding, measured across the board.

The studio’s DFlash draft-model speedup on GLM-4.7-Flash, benchmarked across eleven task domains on partner-lab A100s and reproduced from the run. DFlash lifts throughput well above the MTP baseline, most of all on reasoning-heavy work.

+84%peak speedup · Coding · 103 → 190 tok/s
+51%mean across 11 domains · DFlash w8/v3 against plain decoding
190tok/sat best · 4×A100-40GB · TP=4 · BF16
Coding103 → 190Humanities101 → 146Math104 → 188Multiling.103 → 155QA118 → 157RAG88 → 141Reasoning105 → 175Roleplay100 → 116STEM109 → 163Summ.97 → 133Writing102 → 137

plain decoding DFlash w8/v3 — each cube is ten tokens, emitted at the measured rate. Same seconds; the stream that runs denser is faster.

050100150200190Coding+84 %146Humanities+45 %188Math+81 %155Multiling.+50 %157QA+33 %141RAG+60 %175Reasoning+67 %116Roleplay+16 %163STEM+50 %133Summ.+37 %137Writing+34 %
plainMTP-1DFlash k=3DFlash w8/v3
DFlash on GLM-4.7-Flash, mean output throughput (tok/s) by domain
DomainplainMTP-1DFlash k=3DFlash w8/v3
Coding103139185190
Humanities101131141146
Math104139185188
Multiling.103139155155
QA118145150157
RAG88111139141
Reasoning105138173175
Roleplay100118109116
STEM109139160163
Summ.97120130133
Writing102127130137
plainMTP-1DFlash k=3DFlash w8/v3

sirius · GLM-4.7-Flash on partner-lab A100s (4×A100-40GB, TP=4, BF16) · 80 prompts/domain · mean output throughput, tok/s

A studio that ships systems, not slides.

The whole estate behind the products and the advisory work: 40 repositories across 3 houses, all real, running code across memory, orchestration, containment and infrastructure.

unsigned

platform & agent plane17

plus a small fleet of utilities for deployment, packaging, voice and onboarding

paas

The platform monorepo — Terraform, Helm charts, ArgoCD GitOps, the bare-metal Kubernetes cluster everything runs on.

go · helm · active

Vespertin / os

The agentic OS — Reflex gate plane, signed decision chain, Marrow. Phase one built.

rust · core

Blackwall

AI-containment custody layer — an OS syscall jail with reversible run-custody. v1.0, 500+ tests.

rust · containment

Escapement

Agent-orchestration engine — DAG plans, deterministic dispatch, CRDT merge.

rust · orchestration

Audit and gate plane

The audit & gate plane — agent forge, jury, runner fleet, the record behind the record.

rust · governance

agent-jury

Subagent scoring — an LLM jury (G-Eval + PoLL + ChatEval) with n_eff correction.

rust · quality

herdr

In-house fork of the herdr terminal multiplexer — the agent-session surface.

rust · tooling

lizard

Estate TUI — onboarding and control from one terminal.

rust · tooling

Workload activator

Scale-to-zero workload activator.

go · infrastructure

overflow

GPU-accelerated terminal emulator — wgpu renderer, ConPTY backend.

rust · tooling

Bare-metal substrate

Bare-metal substrate — OS image and provisioning pipeline.

infrastructure

Edge estate as code

The edge estate as code — zones, workers, access, one apply.

ts · infrastructure

Orchestration surface

Live agent-orchestration surface + LLM-gateway catalog.

ts · surface

dc

The research cockpit.

ts · surface

CI runner fleets

Self-hosted CI runner fleets.

infrastructure

Cluster audit

Cluster-audit PWA.

surface

site

The unsigned design canon.

design

cerebral

product & memory plane15

plus the Worker fleet — reverie-cloud · film · files-portal · works · about-worker · soma-brief-worker · gami-vst

Reverie

Neuroscience-grounded memory-consolidation daemon for LLM harnesses. LoCoMo-validated.

rust · active

Cortex

The learning loop and operator console — dispatch, guard, settle surfacing.

rust · active

Cicatrix

Regression-memory + convention-drift framework — the studio learns from its own failures.

rust · memory

Terrarium

Incubator monorepo machine — ideation → execution → validation under enforced gates.

poly · incubation

revenue

The CRM monorepo, built on Attio.

ts · internal

linearctl

Headless Linear CLI — SLSA-attested releases.

tooling

Network policy

Network policy as code, PR-gated.

infrastructure

dreamcode

A Zed fork wiring Reverie behind every agent — the reference integration.

rust · reference

cerebral-design

The design system — tokens, components, the voice in code.

ts · design

Meridian

Fleet-board operations cockpit.

ts · surface

site

cerebral.work — the studio's landing.

astro · surface

voicenotes

Voice-first capture, v1.0 shipped.

product

dealroom / blanklabel

The intake line — deal rooms and white-label deck surfaces, write-only by invariant.

ts · product

ML images

Hardened ML container images.

infrastructure

Inference fleet

Model serving on-cluster and at a partner lab.

infrastructure

the houses & the personal canon

08

vesperan

The holding house's canon — marks, IP, the audit corpus. Counsel-gated.

holdco

zerodays

zerodays.io — the studio's incubation brand and its own site.

site

Harness configuration

The harness-config monorepo — every agent harness, its hooks, skills, and the estate brief's source.

ts · canon

reach

AI-drivable containerized desktop sandbox for agents that need a real machine.

rust · sandbox

pgt

Pretty Good Types — a DSL for types and HTTP routes with full code generation.

rust · tooling

Pact

The promise engine — parked mid-build, revives on demand.

rust · parked

memd

Local caching daemon fronting the memory API.

rust · local

er-terminal

Client for Even Realities G2 glasses — the estate on your face.

scaffold

The full register, house by house, as kept on zerodays.io. Each card is a real repository; the visuals are the register's own.

Signed, tagged, and reproducible from nothing.

The estate keeps a monument register, which we call the gaze: every repository under custody, every commit signed by a known hand, every session on the record.

010,00020,000commits per month02,000agent sessions per month2021-02 · 2 commits2021-03 · 97 commits2021-04 · 293 commits2021-05 · 310 commits2021-06 · 166 commits2021-07 · 220 commits2021-08 · 174 commits2021-09 · 374 commits2021-10 · 145 commits2021-11 · 255 commits2021-12 · 262 commits2022-01 · 354 commits2022-02 · 388 commits2022-03 · 477 commits2022-04 · 493 commits2022-05 · 393 commits2022-06 · 424 commits2022-07 · 563 commits2022-08 · 334 commits2022-09 · 352 commits2022-10 · 623 commits2022-11 · 315 commits2022-12 · 263 commits2023-01 · 321 commits2023-02 · 389 commits2023-03 · 484 commits2023-04 · 461 commits2023-05 · 651 commits2023-06 · 789 commits2023-07 · 601 commits2023-08 · 862 commits2023-09 · 718 commits2023-10 · 1,215 commits2023-11 · 1,368 commits2023-12 · 1,152 commits2024-01 · 1,412 commits2024-02 · 611 commits2024-03 · 619 commits2024-04 · 534 commits2024-05 · 585 commits2024-06 · 471 commits2024-07 · 788 commits2024-08 · 662 commits2024-09 · 533 commits2024-10 · 588 commits2024-11 · 548 commits2024-12 · 479 commits2025-01 · 716 commits2025-02 · 822 commits2025-03 · 935 commits2025-04 · 943 commits2025-05 · 917 commits2025-06 · 779 commits2025-07 · 893 commits2025-07 · 1 sessions2025-08 · 1,020 commits2025-09 · 929 commits2025-10 · 1,080 commits2025-11 · 1,117 commits2025-12 · 2,643 commits2026-01 · 4,044 commits2026-02 · 3,329 commits2026-03 · 3,291 commits2026-03 · 31 sessions2026-04 · 6,974 commits2026-04 · 37 sessions2026-05 · 6,179 commits2026-05 · 535 sessions2026-06 · 8,506 commits2026-06 · 1,801 sessions2026-07 · 10,663 commits2026-07 · 1,793 sessions2026-08 · 6,626 commits2026-08 · 736 sessions2026-09 · 4,539 commits2026-09 · 81 sessions2026-10 · 2,506 commits2026-10 · 372 sessions20222023202420252026peak · 2026-07 · 10,663
monwedfrisun2025-10-01 · 31 commits2025-10-02 · 42 commits2025-10-03 · 32 commits2025-10-04 · 11 commits2025-10-05 · 20 commitsoct2025-10-06 · 62 commits2025-10-07 · 34 commits2025-10-08 · 63 commits2025-10-09 · 46 commits2025-10-10 · 34 commits2025-10-11 · 19 commits2025-10-12 · 23 commits2025-10-13 · 43 commits2025-10-14 · 36 commits2025-10-15 · 34 commits2025-10-16 · 45 commits2025-10-17 · 40 commits2025-10-18 · 16 commits2025-10-19 · 9 commits2025-10-20 · 41 commits2025-10-21 · 52 commits2025-10-22 · 52 commits2025-10-23 · 45 commits2025-10-24 · 21 commits2025-10-25 · 17 commits2025-10-26 · 29 commits2025-10-27 · 39 commits2025-10-28 · 47 commits2025-10-29 · 44 commits2025-10-30 · 29 commits2025-10-31 · 24 commits2025-11-01 · 7 commits2025-11-02 · 13 commitsnov2025-11-03 · 41 commits2025-11-04 · 43 commits2025-11-05 · 39 commits2025-11-06 · 50 commits2025-11-07 · 42 commits2025-11-08 · 10 commits2025-11-09 · 8 commits2025-11-10 · 43 commits2025-11-11 · 85 commits2025-11-12 · 108 commits2025-11-13 · 71 commits2025-11-14 · 26 commits2025-11-15 · 3 commits2025-11-16 · 21 commits2025-11-17 · 46 commits2025-11-18 · 33 commits2025-11-19 · 50 commits2025-11-20 · 68 commits2025-11-21 · 44 commits2025-11-22 · 8 commits2025-11-23 · 8 commits2025-11-24 · 42 commits2025-11-25 · 41 commits2025-11-26 · 49 commits2025-11-27 · 61 commits2025-11-28 · 31 commits2025-11-29 · 23 commits2025-11-30 · 3 commitsdec2025-12-01 · 71 commits2025-12-02 · 103 commits2025-12-03 · 61 commits2025-12-04 · 84 commits2025-12-05 · 86 commits2025-12-06 · 40 commits2025-12-07 · 22 commits2025-12-08 · 75 commits2025-12-09 · 148 commits2025-12-10 · 73 commits2025-12-11 · 56 commits2025-12-12 · 84 commits2025-12-13 · 66 commits2025-12-14 · 24 commits2025-12-15 · 116 commits2025-12-16 · 97 commits2025-12-17 · 120 commits2025-12-18 · 99 commits2025-12-19 · 147 commits2025-12-20 · 69 commits2025-12-21 · 42 commits2025-12-22 · 91 commits2025-12-23 · 20 commits2025-12-24 · 97 commits2025-12-25 · 67 commits2025-12-26 · 71 commits2025-12-27 · 78 commits2025-12-28 · 94 commits2025-12-29 · 135 commits2025-12-30 · 95 commits2025-12-31 · 212 commits2026-01-01 · 172 commits2026-01-02 · 155 commits2026-01-03 · 342 commits2026-01-04 · 119 commitsjan2026-01-05 · 201 commits2026-01-06 · 212 commits2026-01-07 · 180 commits2026-01-08 · 185 commits2026-01-09 · 140 commits2026-01-10 · 100 commits2026-01-11 · 154 commits2026-01-12 · 167 commits2026-01-13 · 104 commits2026-01-14 · 110 commits2026-01-15 · 74 commits2026-01-16 · 132 commits2026-01-17 · 37 commits2026-01-18 · 65 commits2026-01-19 · 121 commits2026-01-20 · 43 commits2026-01-21 · 126 commits2026-01-22 · 92 commits2026-01-23 · 217 commits2026-01-24 · 77 commits2026-01-25 · 116 commits2026-01-26 · 97 commits2026-01-27 · 76 commits2026-01-28 · 160 commits2026-01-29 · 112 commits2026-01-30 · 76 commits2026-01-31 · 82 commits2026-02-01 · 180 commitsfeb2026-02-02 · 119 commits2026-02-03 · 79 commits2026-02-04 · 120 commits2026-02-05 · 207 commits2026-02-06 · 133 commits2026-02-07 · 63 commits2026-02-08 · 21 commits2026-02-09 · 48 commits2026-02-10 · 147 commits2026-02-11 · 93 commits2026-02-12 · 169 commits2026-02-13 · 142 commits2026-02-14 · 89 commits2026-02-15 · 105 commits2026-02-16 · 211 commits2026-02-17 · 87 commits2026-02-18 · 118 commits2026-02-19 · 177 commits2026-02-20 · 136 commits2026-02-21 · 92 commits2026-02-22 · 145 commits2026-02-23 · 107 commits2026-02-24 · 91 commits2026-02-25 · 58 commits2026-02-26 · 133 commits2026-02-27 · 111 commits2026-02-28 · 148 commits2026-03-01 · 142 commitsmar2026-03-02 · 75 commits2026-03-03 · 155 commits2026-03-04 · 138 commits2026-03-05 · 111 commits2026-03-06 · 97 commits2026-03-07 · 27 commits2026-03-08 · 72 commits2026-03-09 · 92 commits2026-03-10 · 122 commits2026-03-11 · 89 commits2026-03-12 · 77 commits2026-03-13 · 125 commits2026-03-14 · 94 commits2026-03-15 · 61 commits2026-03-16 · 97 commits2026-03-17 · 98 commits2026-03-18 · 121 commits2026-03-19 · 97 commits2026-03-20 · 89 commits2026-03-21 · 45 commits2026-03-22 · 108 commits2026-03-23 · 102 commits2026-03-24 · 77 commits2026-03-25 · 73 commits2026-03-26 · 155 commits2026-03-27 · 84 commits2026-03-28 · 127 commits2026-03-29 · 71 commits2026-03-30 · 191 commits2026-03-31 · 279 commits2026-04-01 · 493 commits2026-04-02 · 225 commits2026-04-03 · 216 commits2026-04-04 · 239 commits2026-04-05 · 155 commitsapr2026-04-06 · 255 commits2026-04-07 · 388 commits2026-04-08 · 289 commits2026-04-09 · 315 commits2026-04-10 · 293 commits2026-04-11 · 108 commits2026-04-12 · 87 commits2026-04-13 · 197 commits2026-04-14 · 93 commits2026-04-15 · 130 commits2026-04-16 · 217 commits2026-04-17 · 329 commits2026-04-18 · 264 commits2026-04-19 · 93 commits2026-04-20 · 209 commits2026-04-21 · 248 commits2026-04-22 · 274 commits2026-04-23 · 271 commits2026-04-24 · 342 commits2026-04-25 · 106 commits2026-04-26 · 290 commits2026-04-27 · 194 commits2026-04-28 · 98 commits2026-04-29 · 230 commits2026-04-30 · 326 commits2026-05-01 · 91 commits2026-05-02 · 116 commits2026-05-03 · 135 commitsmay2026-05-04 · 133 commits2026-05-05 · 150 commits2026-05-06 · 302 commits2026-05-07 · 180 commits2026-05-08 · 149 commits2026-05-09 · 184 commits2026-05-10 · 131 commits2026-05-11 · 161 commits2026-05-12 · 286 commits2026-05-13 · 261 commits2026-05-14 · 251 commits2026-05-15 · 585 commits2026-05-16 · 126 commits2026-05-17 · 216 commits2026-05-18 · 58 commits2026-05-19 · 246 commits2026-05-20 · 154 commits2026-05-21 · 219 commits2026-05-22 · 197 commits2026-05-23 · 98 commits2026-05-24 · 71 commits2026-05-25 · 283 commits2026-05-26 · 280 commits2026-05-27 · 301 commits2026-05-28 · 167 commits2026-05-29 · 159 commits2026-05-30 · 183 commits2026-05-31 · 306 commitsjun2026-06-01 · 245 commits2026-06-02 · 326 commits2026-06-03 · 426 commits2026-06-04 · 332 commits2026-06-05 · 337 commits2026-06-06 · 296 commits2026-06-07 · 266 commits2026-06-08 · 417 commits2026-06-09 · 329 commits2026-06-10 · 379 commits2026-06-11 · 222 commits2026-06-12 · 360 commits2026-06-13 · 269 commits2026-06-14 · 346 commits2026-06-15 · 405 commits2026-06-16 · 250 commits2026-06-17 · 271 commits2026-06-18 · 301 commits2026-06-19 · 285 commits2026-06-20 · 182 commits2026-06-21 · 157 commits2026-06-22 · 241 commits2026-06-23 · 288 commits2026-06-24 · 194 commits2026-06-25 · 218 commits2026-06-26 · 279 commits2026-06-27 · 251 commits2026-06-28 · 222 commits2026-06-29 · 172 commits2026-06-30 · 240 commits2026-07-01 · 399 commits2026-07-02 · 465 commits2026-07-03 · 278 commits2026-07-04 · 314 commits2026-07-05 · 330 commitsjul2026-07-06 · 391 commits2026-07-07 · 210 commits2026-07-08 · 358 commits2026-07-09 · 335 commits2026-07-10 · 209 commits2026-07-11 · 508 commits2026-07-12 · 157 commits2026-07-13 · 396 commits2026-07-14 · 519 commits2026-07-15 · 578 commits2026-07-16 · 404 commits2026-07-17 · 337 commits2026-07-18 · 482 commits2026-07-19 · 122 commits2026-07-20 · 106 commits2026-07-21 · 159 commits2026-07-22 · 321 commits2026-07-23 · 375 commits2026-07-24 · 491 commits2026-07-25 · 269 commits2026-07-26 · 290 commits2026-07-27 · 379 commits2026-07-28 · 552 commits2026-07-29 · 297 commits2026-07-30 · 338 commits2026-07-31 · 294 commits2026-08-01 · 258 commits2026-08-02 · 230 commitsaug2026-08-03 · 223 commits2026-08-04 · 322 commits2026-08-05 · 306 commits2026-08-06 · 164 commits2026-08-07 · 313 commits2026-08-08 · 163 commits2026-08-09 · 158 commits2026-08-10 · 142 commits2026-08-11 · 246 commits2026-08-12 · 202 commits2026-08-13 · 329 commits2026-08-14 · 401 commits2026-08-15 · 96 commits2026-08-16 · 481 commits2026-08-17 · 185 commits2026-08-18 · 196 commits2026-08-19 · 449 commits2026-08-20 · 281 commits2026-08-21 · 195 commits2026-08-22 · 88 commits2026-08-23 · 160 commits2026-08-24 · 154 commits2026-08-25 · 273 commits2026-08-26 · 97 commits2026-08-27 · 56 commits2026-08-28 · 98 commits2026-08-29 · 87 commits2026-08-30 · 111 commits2026-08-31 · 162 commits2026-09-01 · 194 commits2026-09-02 · 149 commits2026-09-03 · 114 commits2026-09-04 · 101 commits2026-09-05 · 47 commits2026-09-06 · 34 commitssep2026-09-07 · 135 commits2026-09-08 · 123 commits2026-09-09 · 86 commits2026-09-10 · 56 commits2026-09-11 · 139 commits2026-09-12 · 59 commits2026-09-13 · 120 commits2026-09-14 · 109 commits2026-09-15 · 128 commits2026-09-16 · 106 commits2026-09-17 · 173 commits2026-09-18 · 120 commits2026-09-19 · 179 commits2026-09-20 · 156 commits2026-09-21 · 156 commits2026-09-22 · 78 commits2026-09-23 · 60 commits2026-09-24 · 119 commits2026-09-25 · 76 commits2026-09-26 · 245 commits2026-09-27 · 322 commits2026-09-28 · 655 commits2026-09-29 · 318 commits2026-09-30 · 182 commits2026-10-01 · 202 commits2026-10-02 · 390 commits2026-10-03 · 386 commits2026-10-04 · 567 commitsoct2026-10-05 · 830 commits2026-10-06 · 131 commitsone square per day · size = commits (square-root scale, capped at 401)371 active days of 373

Source · the gaze works register, snapshot of 2026-10-06. Counters exclude nine temporary working copies (git worktrees); the charts count every commit.

Gaze upon the works →

Own the floor, not the spotlight.

Behind the studio stands Vesperan, the holding house of the estate. It owns the marks, the intellectual property and the floor everything else is built on, and it sells nothing.

The estate

holding house
Vesperan

Custody of the marks is the house’s only product: trademarks, copyrights, contracts. It stays quiet on purpose.

operating brands · market-facing
zerodays · unsigned · vespertine

Incubation · AI-compute orchestration · sovereign generative AI.

ventures · incubated
RINA · The Nest · Reverie

Operating companies and products, embedded incubation: one operator, one founder.

the crew
cerebral

An open crew of builders, scientists and operators that acts across the estate, and sometimes serves as the advisory layer for the zerodays incubator.

A product stays a product. Research needs an institution.

Agentic AI in the service of research is a real ambition, and it is not a product feature. Vespertin can serve verticals, even bespoke ones, while a research partner brings infrastructure, tools, broad expertise and co-research. That is why the studio is an institute.

Why an institute

access
Journals, datasets, university agreements

What registered institutions get and companies do not, and the natural home for “infrastructure for acknowledgment” letters.

independence
Scientists who are not officers

The people who measure the claims are not the people who sell them. They decide what science goes public.

method
A need-and-market study first

Who it serves, what it offers and how it gets paid. Nothing is built before the study has answered those questions.

Listen. Watch.

Testimonials, shows and talks, from the people who build it and the institutions that use it. Episodes are in production; these are placeholders for now.

Placeholder cover: A pilot institution, one year in
Video · Testimonial
A pilot institution, one year in

What a signed record changed for the people who sign.

In production · —:—
Placeholder cover: The gate is a commodity. The record is not.
Podcast · Conversation
The gate is a commodity. The record is not.

The studio on why proof, not permission, is the hard part.

In production · —:—
Placeholder cover: Agents that answer to a name
Video · Talk
Agents that answer to a name

Signed decisions for agents, in twenty minutes.

In production · —:—
Placeholder cover: Drift, measured
Podcast · Science
Drift, measured

How an agent's behaviour moves, and how we measure it.

In production · —:—
Placeholder cover: Inside the studio
Video · Show
Inside the studio

A day of agents at work under written rules.

In production · —:—
Placeholder cover: A risk officer on the signature
Video · Testimonial
A risk officer on the signature

Why a named decider matters more than a dashboard.

In production · —:—
Placeholder cover: From collision data to agent behaviour
Podcast · Interview
From collision data to agent behaviour

The science behind the studio's models.

In production · —:—
Placeholder cover: Models a regulator can stand behind
Video · Talk
Models a regulator can stand behind

VesperKinetics: the same discipline, applied to the body.

In production · —:—

Built with the institutions at the frontier.

Compute & cloud

CoreWeave · Lambda Labs · Crusoe · DeepInfra · Vultr · Hetzner

Model labs

Hugging Face · Mistral · Stable Diffusion · NovelAI · GooseAI

Platforms & partners

Anthropic · Alphabet · Meta · Attio

Research & academia

A US national laboratory · UPenn

Contact
the studio

The button sends this to [email protected]. A Cloudflare Turnstile check keeps out spam. If sending fails, your mail app opens instead, with these fields filled in.