NeurionWork with Neurion
Ultra mode
Coordinate demanding tasks through investigation, implementation, and verification.
HydraLabs · Published
Use Ultra for connected work
Ultra organizes a demanding task into work that can be investigated, carried out, and checked. A coordinating agent delegates appropriate parts to managed child agents and brings their results back into the same task.
- Define the result, project or authorized target, and constraints.
- Select a supported model and the Ultra effort or workflow option.
- Review the task setup and any permission requests.
- Follow progress across the investigation, implementation, and verification work.
- Inspect the final files, findings, checks, and remaining limitations.
Use Ultra when a task benefits from several connected steps or independent checks. A short factual question does not require a larger agent workflow.
Keep the same authority boundary
Managed child agents use the task’s selected model profile and applicable authority. They do not receive unrestricted project, browser, or computer access. Delegation does not authorize new targets or consequential external actions.
Work performed by managed agents contributes to the same task and workspace usage. Starting another agent or conversation does not create a separate subscription allowance.
Ultra requires a supported task setup. If the application cannot prepare that workflow, resolve the reported setup issue instead of treating an ordinary run as a completed Ultra run.
CyberBot investigation depth
CyberBot supports quick, standard, and deep investigation depths. Use the scan configuration prepared for the selected workflow and review the target scope before starting. A deeper investigation devotes more work to the approved target.
A deeper scan does not expand the authorized target. Excluded systems, paths, and actions remain excluded at every effort level.
Evaluate the result, not the agent count
More agents and deeper effort do not guarantee correctness. Review the evidence, changes, validation, and stated limits. A successful local workflow check is different from a live evaluation of model quality.
If agents produce conflicting conclusions, use the underlying files and reproducible checks to resolve them. Keep an unresolved issue visible instead of treating consensus alone as proof.