Research · Infrastructure · Products

We research how
intelligence perceives,
reasons, and acts.

Bad Theory Labs is a research lab and product studio. We build the infrastructure that makes AI genuinely useful inside real work — the memory, the retrieval, the context, and the presence that turns intelligence into something that actually follows through.

0%
Hallucination rate
RetainDB · frozen dataset
0%
LongMemEval score
vs 56.7% next best
0
Products shipping
RetainDB · Marrow
Active research · April 2026
The Compression Program
Intelligence is compression. We are investigating whether treating compression as a primary training objective — not a byproduct — produces systems that reason rather than pattern-match.
Persistent context
Selective attention
Native action
Continuous improvement
Intelligence as compression
Memory infrastructure
Causal reasoning
Ambient presence
Persistent context
Selective attention
Native action
Continuous improvement
Intelligence as compression
Memory infrastructure
Causal reasoning
Ambient presence
Our thesis

We research how intelligence
perceives, reasons, and acts
and build the infrastructure
that makes it possible.

What we are

A product and
research lab.

We are not a model company. We are not an AI wrapper. We are building the substrate layer that makes AI genuinely useful inside real workflows — the memory, the retrieval, the context, and the ambient presence that connects intelligence to how people actually work.

Research without product is just paper. Product without research is just imitation. We do both — because neither alone is enough.

How we are structured
01
A research lab
We investigate how intelligence perceives, reasons, and acts. We publish what we find. We build on what we learn.
02
A product studio
We ship. Research without product is just paper. Product without research is just imitation.
03
An infrastructure company
We build the substrate layer — the memory, retrieval, context, and ambient presence that makes AI genuinely useful.
A model company
An AI wrapper
A chatbot
A features team
Another LLM startup
The four capabilities we are building toward
01
01

Persistent
context.

AI systems need structured, reliable memory across sessions, tools, and workflows. Not a bigger prompt window. A real memory layer — one that stores what matters and retrieves it accurately when it matters.

02
02

Selective
attention.

Systems should not surface everything they detect. They need taste, prioritization, and restraint. The goal is not to be helpful all the time — it is to show up only when the signal is high enough.

03
03

Native
action.

AI should not stop at suggestion. It should execute meaningful work. Browse the web, write code, manage files, run tasks end to end. The machine should be able to follow through.

04
04

Continuous
improvement.

The system should become more useful over time by learning the user's patterns, tools, and repeated tasks — without being explicitly programmed to. Intelligence that compounds.

What we are not

Not a model.
Not a wrapper.
Not a chatbot.

The common pattern: bolt AI onto existing software. Optimize for engagement. Ship a demo. Call it an agent. We have no interest in that.

A model company
An AI wrapper
A chatbot
A features team
Another LLM startup
What we are building

The substrate layer
for AI-native computing.

The current generation of AI products is limited in predictable ways. They forget too much. They retrieve the wrong context. They interrupt without judgment. They can generate, but rarely follow through.

A research lab
We investigate how intelligence perceives, reasons, and acts. We publish what we find. We build on what we learn.
A product studio
We ship. Research without product is just paper. Product without research is just imitation.
An infrastructure company
We build the substrate layer — the memory, retrieval, context, and ambient presence that makes AI genuinely useful.

“The goal of Bad Theory Labs is not to ship isolated AI features.
The goal is to help define a new interface to computing.

Bad Theory Labs · Investor Brief · 2025
Products

Infrastructure at the base.
Intelligence at the surface.

Two products. One thesis. Memory and agency — the two things that make AI useful inside real work.

Marrow · Desktop agent

The closest thing
to Jarvis that fits
inside a computer.

Marrow lives on your laptop. It watches what you do, builds up a picture of your work, and stays completely silent — until the moment doing something is actually worth more than the interruption. Then it doesn't suggest. It acts.

01Reads your entire screen in real timeperception
02Builds context silently from everything you domemory
03Decides when stepping in is actually worth itjudgment
04Browses the web, writes code, moves files, uses your appsaction
05Learns your patterns and turns them into toolscompounding
Join the private beta
Marrow · observing
“You've done this three times. Want me to handle it?”
RetainDB · Memory infrastructure

Your AI forgets.
RetainDB fixes that.

Persistent memory and grounded retrieval for AI agents. The highest preference recall on every public benchmark. Zero hallucinations on stored facts. Three lines of code.

Try RetainDB free View benchmark
LongMemEval benchmark · public · reproducible
SystemOverallSingle-sessionHallucination
RetainDB
79%
88%0%
Supermemory
70%
Zep
56.7%
GPT-5 baseline95.5%
Methodology: LongMemEval dataset · GPT-5 + Claude Sonnet 4.5 · temperature 0.0
Hallucination: 16-question code matrix · bleeding-edge SDK APIs · March 2026
Read full benchmark →
Active research program

The Compression
Program

The field has been optimizing the wrong objective for a decade. Next-token prediction induces compression as a side effect. We are investigating what happens when compression is the objective — not a byproduct.

“Intelligence is compression. Not metaphorically — mechanically.”
Read the full program document →
Workstream 01
Compression as objective
What emerges when you train for compression directly? We replace next-token prediction with a genuine MDL objective and study what representations form.
Workstream 02
The reasoning gap
Building an evaluation framework that cleanly separates observational from interventional reasoning — and measuring the gap across current frontier models.
Central hypothesis
Compression → Causal structure → Reasoning
If compression-as-objective produces causally structured representations, and causal structure is what reasoning requires — then compression is the mechanism, and reasoning is what emerges. Falsifiable at every step.
Pre-seed raise

We are raising.
We are selective.

Looking for investors who share conviction about where intelligence is going — not just capital.

Pre-seed · early 2025 · hello@badtheorylabs.com

We are looking for investors who believe

Not a checklist. A real filter. If at least one of these resonates — we should talk.

Memory and context are foundational layers in the agent stack — not features to be bolted on.
Proactive, ambient software will become a major product category in the next five years.
Desktop and workflow-native agents will matter as much as browser agents — maybe more.
Judgment and restraint are underrated differentiators in AI products. Taste is a moat.
A research-driven lab can produce both critical infrastructure and defining end-user experiences.
Get in touch

If this resonates,
let's talk.

We reply to every message. No deck required to start — a paragraph about why you're interested is enough.

Or email directly: hello@badtheorylabs.com

What we need most

Not just capital.
Conviction and alignment.

01
Capital
To accelerate Marrow and RetainDB development, ship a private beta, build the task-aware retrieval API, grow RetainDB's managed cloud tier, and make the first engineering hire.
02
Strategic alignment
Investors who believe in the vision and can provide product conviction, network support, and patience for a large thesis that starts with early signals.
03
Not just capital
The right partner thinks in years, not quarters. We are building infrastructure — the kind that becomes foundational before it becomes obvious.