Black-box AI that takes action without review
Agents that send messages, edit schedules, or update records without preview create unrecoverable mistakes.
Admin assistant
An admin agent that previews everything before it acts.
Yrka's admin agent is a companion dock and full workspace that previews allowed actions, runs permission checks, and waits for human confirmation before writing.
Evaluation focus
AI HR assistant
- AI workforce assistant
- AI tools for workforce management
- review-first AI agent
- AI admin assistant for SMB
- Allowlisted tools — no autonomous tool execution
- Dry-run previews with diff-style change summaries
- Owner-enabled Power Agent rows are explicit and capped
Admin agent · Preview
Admin agent · Preview before write
Build a coverage plan for Thursday PM at Yard A, using the rotating template, and surface anyone past 35 hours.
Dry-run preview · 3 changes
- Assign Jordan P. — Thu 3–11 PM
- Flag Avery T. — projected 39h
- Notify Sam Q. of swap offer
Confirm writesDiscard
Generated layout — synthetic demo data.
Who it is for
- SMB operators evaluating AI agents that can act on workforce data
- Admins who need previews and audit trails before AI writes anything
- Ops leaders comparing review-first AI to black-box automations
- Writes
- Preview-first
Confirm before commit
- Tool registry
- Allowlisted
No silent execution
- Audit
- Every call
Tool-call history
Built for these realities
The pain we hear from operators
Each pain point reflects real conversations with SMB operators evaluating workforce tools today. Yrka is shaped around these realities, not generic feature checklists.
AI answers without source context
Answers that do not cite the data they came from cannot be defended to staff or auditors.
Hidden tool registries
Vendors that will not show you the list of tools an agent can call leave you guessing about scope.
How teams use it
Use cases we ship for today
These are the moments this surface is built around. They map to shipped product behavior and the launch scope for this area.
Companion dock + full workspace
The dock answers scoped questions in context; the workspace handles CSV/TSV action-row artifacts and deeper review.
Ask Agent prefilled prompts
Contextual Ask Agent links prefill source-scoped prompts without running actions, so admins always start from intent.
Confirmed drafts
Source-linked answers and confirmed inbox, schedule, time-off, task, and message drafts give admins a real preview before write.
Why Yrka here
What makes this different
Differentiators map to shipped product behavior. They are the reasons buyers pick this surface after comparing alternatives.
Allowlisted tool registry
The agent can only call tools on an allowlist; permission checks run before previews and before writes.
Thread + tool-call history
Thread history, tool-call history, and rate-limit copy give admins a real audit trail.
Clear AI limits
AI output is reviewed before use and is not legal, HR, payroll, tax, safety, or compliance advice.
How Yrka compares
Honest positioning against the tools buyers usually evaluate alongside Yrka. Comparisons reflect launch scope, not aspirational claims.
vs Generic Copilot/Assistant tools
Permission checks and previews before any write — not a chat sidebar bolted onto an existing app.
How it flows
A workflow that mirrors how operators actually run the day
Three steps, three responsibilities.
- 01
Ask
Use the dock or full workspace for permission-scoped answers and source-linked context.
- 02
Preview
The agent runs the request as a dry run and shows the writes it would perform.
- 03
Confirm
You confirm the writes that should run. Everything is logged in the tool-call history.
What's inside
Admin assistant highlights
Source-linked answers
Inbox, schedule, time-off, tasks, and messages with provenance.
Artifact workspace
CSV and TSV action-row artifacts for review before bulk operations.
Ask Agent links
Contextual links prefill source-scoped prompts without running anything.
Permission checks
Every preview and write checks the same permissions the rest of the product enforces.
Tool-call history
Thread history and tool-call history make agent actions auditable.
Rate limits and refusal
Clear refusal copy and rate-limit signals when something can't run.
Human confirmation, always
The admin assistant is a command surface that previews changes and waits for you to confirm. Owner-enabled Power Agent rows are explicit, capped, and audited.
Honest limits
AI output should be reviewed before use and is not legal, HR, payroll, tax, safety, or compliance advice.
Scope
What's included, what's intentionally out of scope
Yrka maps every feature to its product claim so commercial conversations stay honest.
Included
- Allowlisted admin command surface
- Companion dock and full Agent workspace
- Dry-run previews
- Owner-enabled Power Agent for explicit capped action-matrix rows
- Permission checks
- Human confirmation before writes
- Source provenance for supported read tools
Boundaries
- Autonomous employment, payroll, legal, safety, or compliance decisions
- Broad autonomous model authority
- Unreviewed writes
- Unallowlisted tool execution
Related
Where this fits in the platform
Resources chat
Ask over the company knowledge that's been approved.
Employee and admin Resources chat use organization AI settings and approved prepared source records, with access-scoped retrieval and citations.
ExploreAccess and audit
Permissions, audit events, and records you can actually review.
Role-aware admin surfaces, organization-scoped row level security, audit events, work-item history, and integration ledgers — so the operations team understands what changed and why.
ExploreTrust and security
Real controls. Honest scope. Customer in charge.
Supabase auth, organization scoping, row level security, role checks, private file storage, signed URLs, audit events, AI boundaries, and operations runbooks — described as built, not as promised.
ExploreQuestions
Common buyer questions
We surface the questions buyers actually ask — trial, support, export, mobile, payroll, provider, AI, and trust scope — in plain language.
Will the assistant write to my data without me?
No. The admin agent is preview-and-confirm: it previews allowed actions, checks permissions, and waits for human confirmation before any write. Owner-enabled Power Agent rows are explicit, capped, and audited.
How does the trial work?
Start with a 30-day card-required trial. Cancel anytime from the Stripe customer portal. Workspaces stay available for export or reactivation for 30 days after paid access ends.
Does Yrka process payroll or file taxes?
No. Yrka covers time tracking, scheduling, payroll-prep exports, and provider handoff. Customers stay in charge of payroll processing, tax filing, legal, and HR decisions.
What does mobile look like?
Responsive web and an installable PWA cover employee time entry, scheduling, resources, tasks, messages, and profile review. Employees see scoped offline read context plus queued supported writes with clear sync states.
Can we export our data?
Yes. Payroll-prep exports, schedule and time reports, and employee data exports are part of the core product. Customer data remains exportable for 30 days after paid access ends unless an earlier deletion is requested.
Ready to run the day on Yrka?
Start a 30-day trial in minutes, or book a working session to walk through the rollout with someone who has done this before.