Refresh · Jun 2026
Announcement of trajectories.sh, described as a way to map multi-modal agent paths to a continuous visual timeline.
Evals and environments for computer use and software engineering work
Public company, workplace, funding, and market signals
Updated Jul 29, 2026
Refresh is a YC Spring 2025 startup building high-fidelity simulation environments and evaluation systems for frontier AI labs and enterprises, focused on computer-use and software-engineering agents.
Primary product
High-fidelity RL training environments and evaluation systems for computer-use and software-engineering agents.
Founded
2025
Headquarters
San Francisco, California, United States
Team size
1-10
Industry
Software Development
Sub-industry
AI training data, evaluations, and reinforcement-learning environments
Offices
0 jobs at Refresh
Check back later for new openings
Business model
Stage
Pre-seed
Total raised
$500K
Latest round
Pre-Seed · Jul 2025
Latest amount
$500K
Jul 2025 · Y Combinator
Investors
Research-heavy, realism-obsessed, high-bar, self-directed, and frontier-lab oriented.
Compensation
No public salary bands or compensation details were found; open roles are full-time and San Francisco-based.
Pricing
Custom quote / bespoke dataset and environment engagements.
Differentiators
Technology
Customers
Competitors
Estimated revenue
≈$171k estimated annual revenue (third-party estimate)
Estimated monthly visits
3.5K
Traffic estimate as of Jul 2026
Refresh · Jun 2026
Announcement of trajectories.sh, described as a way to map multi-modal agent paths to a continuous visual timeline.
Refresh · Feb 2026
Announcement of a 4,000-example RLVR dataset in Harbor format, positioned for agentic coding reinforcement learning and reported to improve Terminal Bench pass@1 from 3.4% to 10.1% in their run.
Refresh · Jan 2026
Release of a dataset of 79,139 real developer workflows from 1,669 production repositories, with pytest-verifiable tasks for training production-grade code models.