From our lab to the world
Our team publishes peer-reviewed research in causal AI, graph reasoning, and production simulation, and writes about what we are learning along the way.
Research focus
Institutional Memory
Reasonara is our graph-structured causal memory. It holds over 125M tokens of effective context in production today, at 94% accuracy and 98% fewer tokens than full-context retrieval.
Software World Model
Our Software World Model maps a codebase into a six-layer causal graph: what the code does, why it exists, who owns it, and how a change spreads.
Agent + Human Enterprise
The platform captures decision traces (the reasoning behind every change) and feeds them back to both agents and engineers.
Long Horizon Complex Orchestration
Institutional Memory and the Software World Model let agents hold structured context across long, multi-step tasks.
Publications
Peer-reviewed research
MegaMem
August 22, 2026
MegaMem: A Retrieval Solution for Ultra-Large Context Windows
Source-resolved dual-view retrieval that keeps hundreds of millions of tokens searchable while bounding what reaches the generator. On EnterpriseRAG-Bench (618M tokens), MegaMem improves Overall from 68.22 to 82.26 and reaches 86.50 Correctness.
MEMONDEMAND
August 22, 2026
MEMONDEMAND: A Memory Management System for Large-Scale Enterprise Data
On-demand memory management that builds per-collection hierarchies, separates routing from evidence, and promotes nodes under a bounded budget. It outperforms the strongest published result on EnterpriseRAG-Bench at every scale from 10M to 618M tokens, gaining 12.23% at 10M and 4.66% at 618M.
RCML
May 9, 2026
Multimodal Representation Learning Conditioned on Semantic Relations
Relation-conditioned embeddings yield roughly 31% better Hit@5 than the CLIP backbone across multimodal retrieval benchmarks.
Cielara Reasonara
May 5, 2026
Graph-Structured Causal Memory for Agentic Systems
Enterprise AI fails not because models are weak, but because memory infrastructure is broken. Reasonara holds effective context at 94% accuracy with 98% fewer tokens than full-context retrieval.
SkillLens
May 8, 2026
Adaptive Multi-Granularity Skill Reuse for Cost-Efficient LLM Agents
Skill graphs let agents reuse what they have already worked out. On ALFWorld this lifts success from 45.00% to 51.31% at lower cost.
LARGER
May 8, 2026
Lexically Anchored Repository Graph Exploration and Retrieval
Graph-based codebase navigation that anchors on lexical signals, gaining +13.9 points on localization over retrieval baselines.