Research

A working R&D group, not a marketing department.

We publish the ideas we bet the platform on — foundation model optimization, swarm engineering, cognitive architecture.

All notesModel optimizationCognitive architectureSwarm engineering
Cognitive architectureJan 2026

Visual Fingerprinting for Associative Memory

Compact visual signatures as retrieval keys for long-lived agent memory — associative recall that scales with an agent's tenure, not its context window.

Read the note →
Model optimizationNov 2025

Harness-Level Optimization for Domain Models

Squeezing frontier-adjacent performance out of small on-premise models by engineering the harness — routing, retrieval, and verification around the weights.

Read the note →
Swarm engineeringAug 2025

Swarm Execution & First-Principle Engineering

Decomposing enterprise workflows into agent swarms with verifiable checkpoints — why orchestration, not model size, sets the ceiling on reliability.

Read the note →

Research informs the roadmap.

Every note above shipped as a platform capability. If a problem in your workflow needs new research, that's the kind of engagement we like most.

Bring us a hard problem →