Write-ups
The research runs ahead of the writing.
Every area above runs as working software today; the write-ups trail the work. If one of these areas touches a problem you're carrying, don't wait for a paper.
Research
AI research: memory, agents & code intelligence.
Our products come out of research questions we couldn't leave alone. This is the current slate, every area below is running as working software somewhere in our stack, not sitting in a paper.
Active areas
Four questions we work on daily.
Memory
Persistent AI memory & continuity
What does it take for an AI to genuinely remember, across sessions, across projects, across model generations? We study retrieval that blends fifteen signals, memory that decays and strengthens like the real thing, and continuity layers that let a business switch models without losing its accumulated mind.
Orchestration
Multi-LLM routing & governance
No single model should be a single point of failure. We research routing work across Claude, GPT, Grok, Llama, Cohere, and local models, matching each task to the model that does it best, under governance the owner controls.
Cognition
Cognitive architectures & specialized agents
One generalist agent plateaus; a roster of specialists compounds. We build and measure multi-agent systems, personas with private memory, adversarial reviewers, serial debate chains, and study when a team of minds beats a bigger single one.
Code
Code intelligence at scale
AI that changes code should know the code. We research whole-codebase mapping, every function, class, and dependency across languages, and how that structural truth grounds AI reasoning, from impact analysis to 3D navigation.
Apply it
Research is only interesting when it ships.
Every area above is available as engineering, memory, orchestration, agents, and code intelligence, built into your systems.
Map your workflow →