ACAS — the Autonomous Capital Allocation System — is an AI-native operating environment for research, automation, and software. We design systems that run themselves, prove their own work, and always leave the final decision to a person.
What ACAS is
Most organizations bolt AI onto an existing process. We started from the other direction — building the operating architecture first, then letting products grow inside it. The result is a small team that runs like a much larger one: automated around the clock, documented as it goes, and accountable for every claim it makes.
Data pipelines, monitoring, reporting, and knowledge capture run continuously without a person driving them. The systems watch themselves and speak up in plain language when something needs attention.
Everything we build produces a record. Decisions are logged, results are traceable, and "the model said so" is never treated as proof. If a claim doesn't survive review, it doesn't ship.
No agent moves capital, changes live rules, or alters infrastructure on its own. Automation does the work; people carry the authority. That boundary is architectural, not aspirational.
What we build
ACAS is the map. Each door is its own practice, with its own audience and its own products — all sharing the infrastructure, governance and engineering standard underneath.
Quantitative research and market-intelligence systems: automated data capture, options and equities sentiment analysis, macro context, tagging and scoring of outcomes, and research reporting delivered on a schedule.
Multi-agent systems that carry real operational load — monitoring infrastructure, reviewing work, summarizing what changed, and routing decisions to the people who make them. Delivered as products and as private deployments.
Software for people who make things — audio production tools and utilities that extend the environments creators and knowledge workers already live in, built with the same engineering discipline as everything else here.
We build the thing with you. Agentic solution design, automation of the work nobody wants to do twice, infrastructure and monitoring, data pipelines, and technical strategy for teams adopting AI without losing control of it.
How we work
Anyone can generate an idea now. The hard part is knowing which ones are real. Every substantive thing we build moves through the same gate — and most ideas don't make it to the end. That's the point.
The question, the success bar, and what would count as failure — written down before any work starts.
Implementation against the stated bar, with the reasoning and the data trail captured as we go.
An independent reviewer — human and machine — goes at the work cold, looking for what's wrong with it.
Forward evidence under real conditions. Claims that don't reproduce get withdrawn on the record.
A person decides whether it goes live. Nothing deploys itself, and nothing scales before it's earned.
Ideas that fail their own pre-set bar are retired and recorded, so nobody spends a second week rediscovering a dead end.
When something we said turns out to be wrong, it gets corrected in writing rather than quietly edited away.
Documentation, backups, recovery procedures and redundancy are treated as part of the build, not paperwork after it.
Capability
A working, in-production stack — not a slide deck. This is the shared foundation every ACAS product is built on, described at altitude.
Who we are
ACAS is a small partnership by design. Small enough to move in a day, structured enough that nothing depends on anyone's memory. The automation handles scale; we handle judgment.
Builds and runs the operating layer — servers, pipelines, monitoring, knowledge systems and the automation that keeps them healthy without supervision.
Drives research design, strategy validation and the governance standard every system is measured against.
Get in touch
Whether you want the research, early access to a tool, or a system built with you — start here. We read every message.
Describe the problem and the deadline. We'll tell you honestly whether we're the right fit.
Join the list to receive research notes and be notified when a paid tier opens.
Agentic tools and creator software ship to a small beta group first. Ask to be in it.
Build note — [ALBERT-APPROVE]: create this mailbox on the domain, or swap in the address you want public. Delete this note before launch.