How the Assistant works

The hosted Assistant can investigate eligible Odin and QubeScript tickets and prepare a correction from the exact immutable version linked to each ticket. The ticket language selects its guide, parser and authoring tools: Odin work uses /reported.odin and /candidate.odin; QubeScript work uses /reported.qube and /candidate.qube. Both use the shared interview tools with the matching language engine.

Tickets filed from a running browser test include the recorded interview history when available. Source-load failures and general tickets may have no replayable journey. The Assistant investigates the version and report and may propose a correction.

For Odin tickets, the Assistant can replay a captured path when one is available. Hosted Assistant captured-path replay is not available for QubeScript tickets; it can run fresh tests with the QubeScript runtime. Browser tests support saved journeys for both languages. Replaying earlier steps restores test context; continue through the affected route to check the change.

For a first survey, install the QubeSurvey extension, write and test locally, upload a version for teammate review, activate an open-link Run, then download its data. Follow the four-step survey guide for setup and fieldwork details.

Requesting an investigation

In Tickets, open an eligible, open script ticket and choose Prepare a fix. The Assistant queue selects that ticket. Check the selected tickets, model tier, Maximum cost for this batch and any instructions, then choose Start Assistant and review the confirmation. Only the checked tickets enter the batch; they share one investigation and one maximum charge. Related tickets must use the same source version. Selecting or assigning a ticket does not start work.

Use Back to tickets to return to the report. Assistant run history links directly to its tickets for review. Change assignment is a separate option for assigning responsibility to a person or the Assistant.

For everyday questionnaire authoring, use your own assistant with MCP and the QubeSurvey extension guide.

When requesting Assistant work, you choose a model option and approve a maximum charge. The investigation uses:

  • The exact Odin or QubeScript version linked to the ticket.
  • User notes describing the defect and recorded interview history when available.
  • Surrounding questionnaire logic and structure.

The Assistant does not read ticket attachments. Copy required attachment details into the ticket description or shared instructions before starting work.

Proposals, not replacements

The assistant never overwrites an existing script version or silently alters your source files. When it identifies a solution, it publishes a separate, immutable script version as a proposed fix (or candidate version).

Before using the proposed version:

  1. Compare source: Review the line-by-line differences between the reported script version and the proposed version.
  2. Test the proposed version: Replay available history and inspect where it stops, then answer through the affected and nearby routes. Check corrections with Back where relevant. Start a fresh route with invented answers when replay is unavailable.

Ticket lifecycle and verification

Every investigation leaves an attributable comment on the ticket explaining what was analyzed and what changed. If an investigation fails or produces no fix, the run logs the findings and the ticket remains open for human attention.

flowchart TD
  ticket[Human selects ticket & requests assistant] --> run[Assistant investigates using script context]
  run --> outcome{Outcome}
  outcome -->|Fix proposed| proposed[Publishes immutable version; marks Fixed]
  outcome -->|No fix or failure| record[Records findings; ticket stays open]
  proposed --> review[Human compares source & tests affected routes]
  review -->|Verified| resolve[Human promotes source & resolves ticket]
  review -->|Issue remains| reopen[Human reopens ticket with feedback]
  • Fixed: The assistant marks a ticket Fixed when a proposed version is ready for review. In QubeSurvey, Fixed means ready for human verification, not resolved. The assistant never resolves a ticket or approves its own work.
  • Resolved: Only a human reviewer can verify the result, promote the proposed version into use, and mark the ticket Resolved. If the defect persists, the reviewer reopens the ticket with feedback.

For review and ticket practices, see Tickets and proposed fixes.

AI usage and billing

Assistant provider use is shown as AI usage against the selected Account’s shared balance. Before starting, review the model option, maximum charge and available balance. Reserved credit is released when a run settles. See People, access and billing for Account permissions, prepaid credit and rates.