Monte

Monte is an applied research lab and infrastructure company focused on continual learning and recursive self-improvement for AI agents. It positions itself as a post-training and continual learning layer that helps organizations convert general-purpose foundation models into specialized agents trained on their own work and organizational knowledge. The company frames the problem as today's AI applications being general and static: foundation models do not specialize to a given company's workflows, do not learn from usage, and impose rising token costs for generic intelligence.

The product workflow described on the company's site has four stages. "Capture" extracts training signal from real work, including traces, outcomes, policies, and expert judgment. "Measure" builds evaluations around the workflows, constraints, edge cases, and standards a system must meet in production. "Train" applies reinforcement learning to shape agents around a company's tools, processes, and goals. "Compound" adds memory and routes production outcomes, feedback, and data back into training so models continue improving after deployment. Monte describes its engagement model as research-first and embedded, with its researchers working directly with a customer's team to build evaluation, memory, and post-training systems, and states that the resulting intelligence is owned by the customer.

Y Combinator categorizes Monte under reinforcement learning, B2B, infrastructure, and AI, and lists it as founded in 2025 in San Francisco with a team size of three and Jared Friedman as primary partner. A GitHub organization account under the handle monte-ai hosts a public repository, "expert-augmented-acktr" (Expert-augmented Actor-critic), written in Python.

Founding story

Monte was founded in 2025 by Henry Bae, Alex Fleury, and Jamin Liu. The company describes its team as three Harvard graduates in computer science, statistics, and physics with AI/RL research backgrounds at Harvard, MIT Lincoln Lab, and NASA JPL. Bae holds AB/SB/SM degrees in physics, computer science, and applied math from Harvard; Fleury studied CS, statistics, and AI research at Harvard; Liu studied CS and MCB at Harvard.

Business model

Monte sells to companies as a B2B infrastructure and applied-research provider, combining a platform with embedded researcher engagements that build evaluation, memory, and post-training systems around a customer's real workflows. Sources do not disclose pricing or contract structure.

Not disclosed in the available sources.

Traction

Public sources disclose no customers, revenue, or usage metrics; the only quantitative data points are a team size of three and participation in the YC Summer 2025 batch.

Latest developments

Y Combinator lists Monte as an active Summer 2025 batch company, and the company's website states it is backed by Y Combinator and is engaging with customer teams on evaluation, memory, and post-training systems.