Corepass platform
Bring Corepass models, agent execution and tools into your own products. Build applications that understand tasks, take action and deliver results.
Models
From lightweight specialists to advanced reasoning, our model research explores capability, speed and compute efficiency for real agent work.
Small, fast models for focused tasks such as code search and context retrieval. Specialization helps agents find useful evidence efficiently.
A planned model direction for the most difficult reasoning and agent tasks, exploring deeper reasoning and sustained progress on complex work.
Our main coding and reasoning model direction, with a focus on reinforcement learning for agent tasks and a balance of capability, speed, and cost.
Platform services
Bring model intelligence into a workflow you control, or build on managed agent execution. Two planned interfaces give developers different ways to work with Corepass.
Model access
Connect Corepass models to your application, agent or development framework. Shape context, tools and interaction flows around the experience you want to build.
Managed task execution
Build applications that delegate multi-step work. Agent Harness coordinates models, context, tools and runtime environments, then returns results your application can use.
Tool results inform the next step
Tools & runtime
Bring together the tools, environments and resources an agent needs to act. These capabilities form part of our internal platform foundation.
Give agents a workspace to edit code, install dependencies and run checks and tests.
Let agents work with websites and desktop software, then inspect what changed.
Connect your APIs, data sources and external tools to the same agent workflow.
Make your team’s coding standards and repeatable workflows available on demand.
Keep the relevant instructions, files and progress available across long-running work.
Bring files into the task. Return code changes, reports and outputs your app can use.
Research & engineering
We develop evaluation, training, inference and agent systems together.
The platform grows from the technology behind Remy and our model research.
Data & evaluation
Vera turns real tasks into reproducible environments, agent runs and calibrated verifiers, providing grounded data for training and evaluation.
Model training
Forge connects tasks, model inference, verified feedback and reinforcement learning to support more capable, reliable models.
Model inference
Our serving infrastructure supports model inference and research into faster execution and more efficient compute for agent workloads.
Research into practice
Remy brings models, agent systems and tools into a product people use. It connects our research and engineering with practical coding workflows.
Core capabilities are built internally. We plan to open model access first, then agent execution and tools, guided by business and developer needs. External APIs are not yet open.