Record the model used by the workspace
RunFP can record a local model identity or a Weights & Biases artifact reference and carry that record into captured run evidence. This connects a run to the model metadata available at capture time.
Recording a model reference does not deploy it to a robot or prove the robot actually loaded that file. Keep your deployment process and runtime verification aligned with the record.
Use a local model
Inside the intended initialized workspace:
runfp model set --name nav-policy --version v12 --file policy.pt
runfp model show
Use the real policy file and version for your project. Inspect the result before capturing runs with a task label.
Use a Weights & Biases artifact
Authenticate with your own Weights & Biases account using the installed CLI's login flow:
runfp wandb login
runfp model use ENTITY/PROJECT/COLLECTION:ALIAS
runfp model show
Replace the reference with an artifact your account can access. This integration uses an external service and depends on its credentials and artifact availability. Avoid putting API keys into published command examples or run reports.
Inspect model history
runfp model history
History shows the model record each recorded run carried, with change markers. Pair it with run comparisons to investigate a policy update alongside software and telemetry differences.
Model provenance supports traceability. It does not by itself identify which change caused a regression or provide a complete reproduction environment.
Use the same evidence from an agent
OORB's MCP tools include oorb_model_show and oorb_model_history. Separate model-setting tools change the workspace's model record. Linking a Weights & Biases account remains a person-operated setup action.
Connect your coding agent to inspect existing model and run evidence.