Independent AI research lab

Long-horizon AI systems that improve over time.

r2m.ai builds agents that hold a goal for longer than a context window, remember what they learned, and get better at their own work.

Reason · Memory · Reflect · Improve

01Focus

Four problems that decide whether agents become useful.

Agents that stay coherent across hours, days, and weeks of work — pursuing goals that don't fit in one context window.

Persistent memory that accumulates experience, recalls what matters, and turns an agent's history into its advantage.

Systems that form hypotheses, run experiments, reflect on the results, and refine their own understanding over time.

Long-running autonomy demands predictability — systems that are steerable, auditable, and safe to trust with real work.

02Research

An autonomous research loop that can't grade its own homework — literature-grounded hypotheses, real experiments, frozen scoring.

ML price models, 101-Alphas factors, options strategies, and guarded live execution — one auditable decision loop.

Six sources, one briefing — the day's AI research collected, distilled, and delivered before you wake up.

03Team

We came from the labs that built frontier models and the desks that had to trust them with money.

Industry

Academia

We're a small team working on agents that last — and we're looking for people who want to build them with us.