Our mission
Put AI to work for every person and every team.
Forward Deployed Engineers bring agents into real enterprise workflows. Enterprise RSI keeps improvement measured and controlled. Expert data, agent trajectories, evaluations, and RL tasks strengthen the learning loop behind both.
Deployment and learning belong in the same loop. FDE grounds the work in an enterprise’s systems and constraints. Enterprise RSI makes improvement repeatable, with evaluations, approval gates, and rollback. The training gym supplies expert-created challenges and feedback to strengthen that loop.
Why “Seatrial”
A sea trial puts a vessel through its paces in real conditions. Performance is observed and measured, beyond what the drawings promise. The name reflects our belief that capability should be demonstrated through the work.
For agents, that means reliable deployment, verifiable outcomes, and controlled opportunities to improve. Our mark is the Plimsoll line: a vessel’s load line and a reminder to understand the conditions in which a system can perform.
How we work
Start with real work
Embed with the people who own the workflow. Build the integrations, evaluations, and operating practices that let an agent do useful work in their environment.
Measure before promoting
Agree on the baseline and success criteria. Evaluate proposed changes against held-out tasks and regression checks before making them the new baseline.
Keep improvement bounded
Enterprise RSI follows a defined cycle: propose, evaluate, gate, and promote. Set the scope, iteration limits, and change budget; keep evaluation criteria and approval gates outside the loop's reach.
Keep people in control
Route consequential changes to a human for approval, record the decision, and preserve a known-good version for rollback. Promotion is a controlled release decision.
Use data with permission
Agree on data rights and workflow boundaries before using traces for training or evaluation. Customer workflows are not a shared training asset by default.
Make the next cycle useful
Turn observed failure modes into expert-created tasks, trajectories, and verifiable feedback. Leave your team able to operate the system and understand why a change was accepted or rejected.
Three connected capabilities
Forward Deployed Engineering
Embed with your team to integrate agents into real workflows, build evaluation and reliability infrastructure, and hand over systems your team can operate.
Explore FDE →Enterprise RSI
Bounded recursive self-improvement: propose changes, evaluate against a baseline, apply approval gates, and promote accepted versions with a path to rollback.
Explore Enterprise RSI →A training gym for agents
Build expert data, execution trajectories, evaluations, and RL tasks around capability gaps, using permission-cleared workflows and verifiable outcomes.
Explore training →Our long-term direction
From deployed agents to continuous learning.
FDE and Enterprise RSI form the foundation. Our ambition is to deepen the training capabilities around them: model improvement infrastructure, interactive training environments, and eventually infrastructure for learning from experience. These are future milestones, grounded in permission-cleared data and measurable progress at every step.
Explore the roadmap →What should your agents be able to do next?
Start with a workflow to deploy, an improvement loop to build, or a capability gap to close.