Deputy

Integrations reach across several system classes and write into them. Deputy states nearly every piece of the application is available for integration, and publishes developer guides for three classes (point of sale, payroll and HR) alongside a REST surface for writing employees, timesheets, leave, availability, budgets, locations and sales metrics.

The catalog at deputy.com/integrations covers payroll and accounting (ADP, Paychex, Gusto, Xero, QuickBooks, MYOB), point of sale (Square) and identity (Okta, Azure, Oracle through SSO). Zapier, Workato and a Singer tap let a customer add their own.

The integrations act. Hours captured in Deputy are pushed into the payroll system as authenticated writes rather than exported for someone to type in again, and demand signals are pulled from point of sale systems to drive scheduling.

Sourcedeveloper.deputy.com and deputy.com/integrationsread 2026-09-08

Deputy's agent works beside its rule engine, not through it. The Labor Model appears only in the developer documentation, as a rule engine addressed by API with operations to create a rule, list rules, update a model and run a model, and DeXML is a server side scripting language per install in the same place. The AI page mentions neither.

It describes Deputy AI as a conversational agent that drafts a schedule from recent patterns, finds a replacement across several days, extends shifts, updates availability and approves timesheets, each on explicit confirmation. So the runtime is a deterministic engine that API callers drive, and the agent does its own work beside it: one agent taking many steps, not a named runtime the agent orchestrates.

Work does run across participants, with shift offers taken up by workers, roster swaps that need a counterparty, task groups assigning across a team, and approvals routing to a manager. Scheduling connecting to time, attendance and payroll is data flow, not orchestration.

Sourcedeputy.com/ai and developer.deputy.com Smart Scheduling guidesread 2026-09-08

The agent grounds in a maintained corpus that persists between runs. Employee records, availability, qualifications, training and role requirements, historical demand, wage and labor budgets, and an encoded labor law and pay rules layer all live in the platform, kept current by continuous HRIS synchronization, and are read on every request.

Deputy AI follows the same rules the team already uses and weighs availability, leave, shift conflicts and role or training requirements when preparing an action. The encoded compliance layer spanning multiple jurisdictions persists, scales past any context window across a whole workforce, and stays queryable.

No index, graph or embeddings layer is documented.

Sourcedeputy.com/ai and developer.deputy.comread 2026-09-08

Deputy AI cannot finalize anything without the user's confirmation, and the product shows it. The AI page's FAQ states that Deputy AI follows a human in the loop approach and cannot publish a schedule or approve a timesheet without explicit confirmation, showing a summary of the action it is about to take, with nothing finalized unless the user confirms.

The product surface matches: a tap to confirm step where every action comes with a clear plan before it goes ahead, voice actions on mobile with every action ready to review first, and relevant warnings surfaced in the preview so the manager can decide before anything is updated.

The review and approve surface is Deputy's own, built over the agent's actions. Scope is bounded by the role based permissions the account already uses: the agent can only act where the user could.

Sourcedeputy.com/airead 2026-09-08

On compliance, Deputy is ISO 27001 certified, UK Cyber Essentials certified and SOC 2 Type II compliant, handles PCI-DSS as a level 3 merchant on an annual SAQ A self assessment with quarterly scans by an Approved Scanning Vendor, and is listed with the Cloud Security Alliance at CSA STAR Level 1. Certificates, attestation reports and security documentation can be downloaded from the trust center at deputy.com/trust.

On access, corporate customers get single sign on through Okta, Azure and Oracle, multi factor authentication is available on every account, and customizable role based access levels show employees, managers and administrators only what is relevant to them. The agent inherits this: Deputy AI respects the role based permissions already configured in the account and can only show and act on what that user already can.

ISO 27001 and Cyber Essentials are certifications, SOC 2 Type II is stated as compliance, and PCI-DSS is a self assessment.

Sourcedeputy.com/security and deputy.com/airead 2026-09-08

Changes inside the platform are recorded, but not what the agent did or why. The Resource API exposes change history and journal objects, among them EmployeeHistory, EmployeeAgreementHistory, Journal, SmsLog and SystemUsageTracking, so Deputy supplies a retrievable record of changes inside its own system.

Projected labor cost while building schedules, GPS verified time tracking, adherence and analytics dashboards measure the customer's workforce, not the agent.

Nothing documents an agent run trace, a decision record, or attribution separating a change made by Deputy AI from one made by a manager, so a buyer cannot reconstruct why the agent did what it did.

Sourcedeveloper.deputy.com/docs/getting-started-with-the-deputy-apiread 2026-09-08

What is documented is a chat surface. Deputy AI is a conversational agent operating through chat, on the web and in the mobile app by text or voice, with core scheduling and timesheet actions completed via chat. Conversation state within a session is what that surface implies.

Persistent records of employee availability, qualifications and historical demand are business records, not agent memory.

No memory layer with a stated scope or lifetime, per user, team or workflow, is documented, and no retention or deletion control over agent memory is published. Nothing published says what is kept between turns.

Sourcedeputy.com/airead 2026-09-08

A named region list is offered against a stated sovereignty need. The security page says the infrastructure is ready to support the customer's data sovereignty needs in three strategic regions (Australia, the United Kingdom and the United States), with services hosted in one of the AWS regions in those countries and customer data kept logically separate through sharded database partitions and deployment across regions. The regions are offered to the buyer, not just a fact about where the vendor happens to run.

Operating in more than eighty countries is market coverage, not hosting. No VPC, on premises or customer tenant option is documented.

Sourcedeputy.com/securityread 2026-09-08

Deputy AI is one agent, not a set of agents a buyer selects among. The FAQ answers the plural question directly and describes a single assistant for admins and managers, scoped by their role.

What is adoptable is a browsable AI prompt library, published and linked from the AI page, with named entries shown inline: real time floor view, build a schedule from past patterns, peak demand cover, leave and availability, payrun readiness check, and multi day sick cover. Alongside it sit industry configurations for retail, hospitality and healthcare, regional compliance rule sets and configurable labor model rules.

Those are ways of asking one agent to do things, not separate products a customer adopts on their own.

Sourcedeputy.com/airead 2026-09-08

The platform publishes a real event surface. Webhooks, set up inside the Deputy install, deliver data as soon as an update is made rather than waiting to be polled, with a published webhook action list, worked timesheet and new employee examples, manual webhook configuration, a Webhook resource object, and AWS SQS as a queue target. Zapier and Workato add event routing without code.

Channel coverage is wide: web, the mobile app, chat, voice actions in the worker's own language, SMS with its own SmsLog resource, and push notifications for schedule changes, open shifts and swap approvals. The AI layer flags coverage gaps and schedule issues ahead of the shift.

Deputy AI itself is invoked by a manager rather than starting on its own, so the event surface reaches the platform and the customer's integrations.

Sourcedeveloper.deputy.com/docs/webhook-overview and deputy.com/airead 2026-09-08

The customer does not choose the model. No model picker, admin model setting or routing control appears on the AI page, the security page or the developer documentation, and the AI FAQ discusses the LLMs that power Deputy AI without offering a choice. Deputy's newsroom states the platform was built with Amazon Web Services generative AI services, a disclosed provider, and the security page names AWS as the cloud host. A named single provider makes the absence of choice visible.

The vendor states that customer information is not used to train the models and that customer data is not processed by the LLM itself. That is a data handling boundary, not a choice of model.

Sourcedeputy.com/ai and news.deputy.comread 2026-09-08

Deputy publishes a deep, open developer surface. It describes itself as "the API-first workforce management platform" with nearly every piece of the application available for integration: an ungated REST API on v1 and v2 addressed at the customer's own subdomain, OAuth 2.0 or permanent token authentication, an API reference that can be tried in the browser, a Postman collection, a changelog, an OpenAPI index at developer.deputy.com/llms.txt, and roughly sixty documented resource objects spanning employees, timesheets, rosters, leave, pay rules, tasks and webhooks.

Beyond REST there is Deputy Embed, with a provisioning API, calls made on behalf of an embedded account and an iFrame embed for putting Deputy inside a partner's product; a Singer tap; Zapier and Workato; and DeXML, a server side scripting language per install. These are outside callers driving the platform. No public agent SDK or MCP server is documented.

Sourcedeveloper.deputy.com/docs/getting-started-with-the-deputy-apiread 2026-09-08

Nothing documents testing the agent. Across the AI page, the security page and the developer documentation, there is no evaluation harness, scored test cases, simulation surface, quality gate in a release path, or optimization loop run against the agent.

Previewing projected labor cost and validating compliance before publishing a schedule gates each roster, not a change under test. Demand forecasting from point of sale data measures demand, not the agent. Running a Labor Model executes a rule set rather than testing one. The vendor's accuracy and time saving claims are marketing.

Sourcedeputy.com/ai and developer.deputy.comread 2026-09-08

No hosted or local browser, desktop session, or remote or local computer control is documented on the AI page, the security page or the developer documentation. Deputy AI acts inside Deputy's own workflows through its own APIs, and the platform reaches outside systems through a REST API, webhooks and an integration catalog, all programmatic.

Two features come close without being computer control. Voice actions on mobile let a manager speak a change instead of typing it, which is an input channel to the agent. The time clock Kiosk, which has its own API resource, is a purpose built device app the vendor controls, not an agent driving software it does not control.

Sourcedeputy.com/ai and developer.deputy.comread 2026-09-08