日本語 
FlashLabs has introduced a new capability for OrcaRouter that allows AI-agent execution records to be transferred between local environments and the cloud. The feature uses two commands, “orca push” and “orca pull,” to move execution records in either direction, making it easier for developers to inspect, reproduce, and compare AI-agent runs.
The system uses a common format called orca-trace v0, combining information from local agent activity with cloud-side model interactions. Local records can include commands, file changes, and external tool usage, while cloud records capture model requests and responses, model routing, and automated communications.
The ability can help with fixing problems and checking how well things work by letting programmers look at how an agent worked without needing it to be running. It also lets them see how different AI models act in the situation in a process. Before moving these records from one place to another meant people had to download them using a web browser. Then upload them again.
Also Read: Uravation launches local LLM support for companies with strict AI security rules
This new feature shows that there is a need, for being able to watch what is happening and make sure things can be done again in AI that works on its own. As AI agents handle complicated jobs that use tools, files and several models keeping good records of how they work can help teams find out what went wrong check how the models act and make testing more reliable.
FlashLabs has also incorporated safeguards around transfers, including destination confirmation, dedicated keys, integrity checks, and secret-information inspection.
English