Intelligence that
doesn't just suggest.
It reads. It acts.
Kumo HR Agent 2.0 isn't a copilot bolted onto an HRIS. It's an agent colleague that reads the document, sees what the last tool actually returned, and decides the next move. Give it a goal: it plans the mission, runs every step, pauses for your confirm on anything sensitive, and files the result as an artifact, with a full log, and an honesty kernel that cannot claim unfinished work.
It reads. It sees.
It keeps going.
Copilots suggest. Kumo HR Agent 2.0 is agentic: it watches every tool come back, then takes the next step. It reads CVs, contracts, and policies with a steely eye, tells you honestly what it did, and still waits for your confirm on anything that matters.
Kumo HR Agent 2.0 vs a typical HR copilot
| Capability | Typical HR copilot vs Kumo HR Agent 2.0 |
|---|---|
| Executes the missionNot a suggested to-do list | Suggests Runs every step |
| Loops on live resultsSees what the last tool actually returned | One-shot Keeps going |
| Reads CVs, contracts, policiesFirst-pass, with receipts, not legal advice | Summarises Reads & files |
| Confirm-gated & honestCannot claim work it has not finished | Sounds done Shows the log |
Qualitative agentic index: how far each system goes on the same HR mission. Not a third-party lab score.
Missions. Artifacts. Skills.
Most "AI for HR" is a search bar with a thesaurus. Kumo works like a colleague: it takes missions, produces artifacts, and learns your org's skills. Three pillars, one operating system.
Missions
Give Kumo a goal and it plans the steps, runs them, and pauses for your confirm on anything sensitive. 2.0 sees each tool result and keeps going.
- "create me a hiring plan" → 6 steps, run
- Adjusts the plan as it learns from live results
- Confirm gates on sensitive steps
- Full log of every step & tool call
Artifacts
Everything the agent produces is saved, versioned and reusable: hiring plans, job posts, audits, reports, policies.
- Every mission ends in an artifact
- Versioned, searchable, shareable
- Reuse as the start of the next mission
- Yours to edit, by hand or by ask
Skills
Teach Kumo how your org runs a process once, and it does it that way every time, for everyone.
- Encode your playbooks as skills
- Consistent across every manager
- Skills compose into bigger missions
- Update once, applied everywhere
A conversation with your whole org.
Type like you'd message a colleague. Kumo looks across hiring, performance, payroll, leave and policies, plans the mission, and gets to work.
Cross-module questions
"Who are the engineers in EMEA without a stretch goal this quarter?"
Drafts & documents
"Draft a parental leave policy update for Germany, in German."
Operational nudges
"Run a pay-equity check across product, flag anyone > 8% off-band."
Live reporting
"Compare attrition in Sales vs. last quarter, by tenure band."
See what's coming, not what's gone.
Kumo AI scans every event across your org · promotions, comp, pulse, calendar load, hiring throughput · and surfaces what needs attention now, with the receipts.
Signals · this week
12 newRetention risk · Sales EMEA
Pulse score dropped 0.4 over 3 weeks. 5 reports flagged. Avg tenure 3.8y.
Compliance · UK IR35 status
5 contractors hit 12-month threshold. Recommend in-house conversion review.
Hiring pipeline · Engineering
Time-to-offer up 11 days. Conversion holding, but offer-acceptance dropped 7%.
Pay equity · Product
Adjustment plan approved by Maria & Tom. Live in next pay cycle (21 May).
Skill gap · AI engineering
3 product squads understaffed for upcoming roadmap. Suggested: 2 hires + 4 internal moves.
A predictive layer over every module.
Signals draw from every event in your people graph, not just one module. So a drop in pulse plus a comp outlier plus a manager change becomes one alert, not three Slack threads.
Agents that finish the job.
Built-in workflows that go beyond drafting. Each agent is scoped, observable, and revocable, with an immutable audit trail of every step it takes.
From offer to signature, in minutes.
Drafts a localised contract from a hiring decision, routes for review, e-signs, files to People, and triggers onboarding.
Reviews that actually close on time.
Sends nudges, summarises peer feedback across 3 sources, drafts a manager review and surfaces calibration risk.
Legislation tracked. Policies, refreshed.
Watches statute changes across 180+ jurisdictions, re-drafts impacted policies, publishes to the handbook, in four languages.
Pay equity, on a schedule.
Runs equity checks, flags drift, models the cheapest path to compliance and prepares manager-tone adjustment notes.
Day-one ready. Day-90 thriving.
Generates the onboarding plan from role + locale, books equipment, assigns a buddy, and books the welcome session.
Engagement, but actually actionable.
Designs the right survey, runs it, themes the comments, and routes findings to the right manager with a 1:1 prompt.
See what it can finish today.
Not a feature checklist. The same registry that gates what the agent may execute. Browse by module: leave, rota, hiring, pay, reviews, and more. Each entry shows how to ask, who can use it, and the confirmation that protects it. Coverage grows with every release.
- “Clock me in”
- “Build next week from the morning template”
- “Book leave for Priya next Thursday”
- “Draft a job post for the Berlin AE role”
- “Run a pay-equity check across Product”
- “Chase overdue reviews before Friday”
Works with your AI stack.
Embed Kumo's complete model and durable agent loop in your own product, or let Claude, Cursor, and your agents call Kumo over MCP. It works the other way too: connect any system you already run, including the old ones, and Kumo acts in it behind approval. Every model call, decision, tool call, receipt, and outcome is traceable.
- Claude and Claude Code
- Cursor and VS Code
- Your own agents, over REST or MCP
- Your product, with the Kumo Agent SDK
- Calendars, chat, and internal tools
- Legacy systems, through a connected adapter
Powerful by default. Safe by design.
Kumo AI holds the most sensitive data in your company. We built it . With controls designed for regulated industries and reviewed by independent auditors.
Built on your data. Never trained on it.
Customer data is isolated and used only to answer your queries. Models are never trained on your tenant's content.
Permission-aware retrieval
The AI sees only what the asking user is allowed to see, down to the field. RBAC isn't a filter, it's a barrier.
Cited & explainable
Every answer cites the records it used. Every action explains its reasoning in plain language. No black boxes.
Immutable audit log
Every prompt, retrieval, action and approval is logged. Searchable, exportable, retention configurable per region.
Data residency & isolation
EU, UK, US and APAC regions. Customer data never crosses regions without explicit consent. Tenant-isolated by default.
Reversible by default
Kumo's own actions can be reverted from the run log. Critical actions require explicit human approval every time, and anything running in a system outside Kumo always does.
Give your people team a force multiplier.
30-minute demo · live data · your real questions.