Outhad AI Research Lab
Applied intelligence for the live session.
Outhad AI is a research and product lab. We build systems that decide in milliseconds — and the orchestration layer that puts them to work.
Convertive
Real-time customer activation for ecommerce. Predicts intent from live behavior and intervenes during the session — before an anonymous visitor leaves.
Outhad Studio
An orchestration layer and builder for agent systems. Compose, deploy, and supervise multi-agent workflows without rebuilding the plumbing each time.
KoraCode
An open-source agentic harness for the browser. Describe a web task from the terminal and it drives a real browser — step by step, in view, and asking before anything irreversible.
Outhad ContextKit
An open-source agentic memory layer. Agents carry persistent context across users and sessions — a temporal, causal knowledge graph that recalls only what matters, with PII encryption built in.
Streaming decisions & RL
Decisioning under hard latency budgets: models that have to answer in milliseconds, improved by a continuous optimization loop rather than a nightly batch.
Thesis
Most of the stack acts too late. Behavior is captured in real time and then processed on a schedule — hours or days after the moment that mattered. By the time a decision is made, the session is over and the visitor is somewhere else.
We think the interesting problems are not about raw capability. They are about making capability useful inside real constraints: a few hundred milliseconds, a visitor with no history, and a cost budget that has to survive contact with production traffic.
We are building in two directions at once.
- Decisioning under real cost and latency constraints
- Agent systems that can be composed, supervised, and shipped
One is a product that has to work on live commerce traffic today. The other is the layer that makes those systems composable for the next problem.
We are a small team of applied researchers and engineers.