AI features currently available
StaffOps currently has two user-initiated AI tools: the AI people extractor, which proposes a person profile from selected documents, and the AI assignment assistant, which proposes an assignment draft from instructions and optional files. Shared files can be selected as input, but nothing is sent until an authorized user starts the relevant AI action.
These tools structure information for review. They do not autonomously hire, reject, deploy, schedule, discipline, pay, score or monitor a person.
Current provider and request settings
The current server implementation sends AI requests to the OpenAI Responses API. It is presently configured to use an OpenAI GPT-5.6 model with structured JSON output and sends store: false on each request. StaffOps may change the model or provider as the product evolves, but this page should be updated when the data flow materially changes.
What the people extractor sends
For a people extraction, StaffOps sends the selected PDF, DOC/DOCX, text, CSV or image files; optional operator notes; the organization’s approved operational location directory; and the existing saved values used by candidate dropdown fields. Images are sent as image input and other supported files as file input.
The prompt instructs the model not to invent missing facts, to leave uncertain values blank, and to choose existing organization options rather than creating new ones. The model can propose a profile photo only from a separately uploaded image file.
What the assignment assistant sends
For an assignment draft, StaffOps sends the operator’s instructions and selected files, plus organization context needed to match the draft to existing records: active candidate names, client names, contract numbers, service names, approved location codes and saved routes.
The model is instructed to use only close matches from those directories, leave unmatched values blank and avoid inventing a location, route, candidate, client, contract or service.
File staging and retention
AI source files are first uploaded to the organization’s private Firebase Storage intake path so the server can validate and read them. The current limit is six files, no more than 10 MB each and 25 MB combined.
If the user approves the result, the intake files become source documents attached to the created person or assignment. If the user cancels or the request fails, the app attempts to delete the staged files. A network or service interruption can delay that cleanup, so administrators should contact Support if sensitive intake content appears to remain unexpectedly.
Review before anything is created
The people extractor displays the proposed profile, summary and warnings and creates the person record only after the user selects Approve & create. The assignment assistant displays a proposed assignment, requires a saved route where origin and destination are present, and creates only a draft after approval.
Created assignment drafts are marked as needing review. Users must verify dates, times, client, contract, service, route, people, instructions and source documents before operational use or billing.
No automated high-impact decisions
Do not use StaffOps AI output as the sole or determinative basis for hiring, rejection, deployment, scheduling, discipline, compensation, access, safety, performance evaluation or another decision that materially affects a person. A qualified person must review the source material, correct the output, apply the organization’s policies and make the final decision.
Accuracy and prohibited inputs
AI output may be incomplete, inaccurate, outdated or unsuitable even when it follows the requested schema. Users remain responsible for checking it against the original material and deleting incorrect information.
Use the minimum information necessary. Do not submit passwords, authentication codes, payment-card or bank credentials, secrets, unrelated government identifiers, medical information or other highly sensitive data unless the organization has expressly approved the use and completed the required legal, privacy and security review.
Access and organizational controls
AI actions require an authenticated, active organization member and are exposed according to the user’s ability to create people or assignments. Workspace owners and administrators should decide who receives those permissions, train users on review requirements and monitor compliance with local employment, privacy and AI rules.