IntelligenceCCoommpprreessssiioonn

We build deep research agents as a path to structured world models. Real machine intelligence means solving compression.

Defining the future of World Models

Read our thesis on Compression
INTSUM
PRODUCT LAUNCH

Introducing Intsum

A deep research decision support agent. Intsum ingests thousands of news articles, social posts, and documents daily and provides users with an Intelligence Summary to aid decision-making. It's like your own presidential level daily brief.

Try Intsum (Open Agent-v0)
9:41
Intsum AI

Current Task

Analyzing 152 Papers

Reasoning...
Found a correlation between entropy reduction and latent space capability in 42 papers.
Generating summary of "Learning World Models from Ray Tracing"...
Ask a follow up...

Research Core

v1.0.4 · Active

Context10M Tok
Sources50k+
View Demo

Proactive Decision Support

Intsum is a proactive deep research agent designed to maintain critical context across your daily workflows.

Powered by our Autonomous Research Engine, it runs continuously in the background—performing literature search, data analysis, and hypothesis generation.

It doesn't just answer questions; it anticipates decision points, synthesizing fragmented updates into coherent, actionable intelligence.

class Intsum(DecisionAgent):

def __init__(self):

self.memory = PersistentContext()

self.engine = ResearchEngine( mode="autonomous")

async def monitor_workflow(self, user_state):

# Proactively bridge knowledge gaps

current_context = self.memory.retrieve(user_state)

gaps = self.engine.identify_blindspots(current_context)

if gaps.is_critical():

# Launch deep research cycle

insight = await self.engine.deep_dive(gaps)

self.memory.update(insight)

return insight.synthesize()

Active Context Monitor

Scanning Decision Landscape

Context RetentionCross-session memory
Continuous
Deep ResearchMulti-step reasoning cycles
Autonomous
Proactive AlertsBlindspot detection
Live
05:40
T

Research

Read our latest papers on world models, entropy reduction, and decision support systems.

Read the papers (Coming soon)
active
Open PR

Open Source

We believe in open science. Our core compression algorithms are available for the community to study and improve.

Explore our research on GitHub or contribute to the next generation of world models.

Join us in building something you can never finish.

Get early access updates and express your interest in joining the lab. We review every submission and invite promising builders to an on-site interview.