Build agents in plain language, or start from a proven template.
Describe the job and Kizen assembles the steps, or start from an agent that already fits. Connect the systems it needs, even ones without a connector.
Ready-made agents for classification, extraction, voice, meetings, integrations and compliance
Describe the task in plain language; Kizen builds the workflow steps
Connect your systems, with code steps for anything that has an API
Choose from 50+ AI models for each step
Claims Intake AgentDraft
TriggerNew email to claims inbox
AIClassify document
AIExtract policy no., loss date, claimant
GuardrailRedact PII · check policy rules
HumanAdjuster approves
Code stepUpdate claims system
02Test
Prove it works on your real work before it touches production.
Run the agent on your own documents, calls and cases. See exactly what it did at every step, tighten the guardrails, and get the business to sign off before launch.
Run agents on real samples of your documents, calls and cases
Inspect every step's inputs, outputs and decisions
Set guardrails like PII redaction, policy rules and scope limits
Get sign-off from the business before go-live
Test run · sample claimsComplete
claim_0918.pdfPassed
claim_0919.pdfPassed
claim_0920_scan.jpgNeeds review
claim_0921.pdfPassed
fnol_call_0922.mp3Passed
Every step's inputs and outputs are saved for review.
03Deploy
Launch into the workflows and apps your teams already use.
Agents go live inside your real processes, with a person in the loop wherever the stakes call for it and access limited to exactly what each agent should see.
Run agents on triggers, on a schedule or on demand
Human approval on any step you choose
Role-based permissions, so agents only see the data they should
Runs on ISO 27001, SOC 2 Type II and HIPAA-ready infrastructure
Claims Intake AgentRun #4812Running
Classify documentclaim_0921.pdf → First Notice of Loss
Extract fields
Policy no.
[POLICY #]
Loss date
[DATE]
Claimant
[NAME]
Guardrail checkPII redacted · policy rules passed · within scope
Human reviewRouted to claims adjuster · awaiting approval
Every step logged to the audit trailRejectApprove
04Monitor
See every run, every decision and every exception.
Know what your agents are doing at all times. Every action is logged, exceptions surface for review, and the right people hear about anything that needs attention.
A full audit trail of every action and decision
A review queue for anything waiting on a person
Track volume, exceptions and turnaround over time
Notifications when something needs attention
Claims Intake Agent · this weekLive
1,284Runs
12Awaiting review
3Exceptions
09:42 · Claim routed to adjusterLogged
09:40 · PII redacted (2 fields)Logged
09:37 · Missing policy no.Exception
05Optimize
Get better every week, not just on launch day.
Use what reviewers correct and what goes to exceptions to make the agent sharper. Swap in better models as they launch, and automate more steps as trust grows.
Turn reviewer corrections and exceptions into better instructions and rules
Switch to better models as they launch, without rebuilding the workflow
Move steps from human approval to automation as confidence grows
Measure the impact, from hours saved to faster turnaround
Change historyv3
v3Switched extraction to a newer model
v2New rule: flag claims missing a policy number
v2Auto-approve routine address changes
v1Launched with adjuster approval on every claim
Start faster
Popular agents saving time and improving experiences.
Document Classification
Sort every inbound file — applications, claims, statements — the moment it lands, and route it to the right queue.
Reviews every recorded call, email and message against your best practices and regulatory rules, and flags potential risks before they become problems.
How do you keep AI agents accurate and in control?
Every agent runs inside guardrails you set: human approval on any step you choose, checks like PII redaction and policy rules, and access limited to the data each role should see. Every action and decision is logged to a full audit trail, so you can see exactly what happened and why.
Which AI models does Kizen work with?
Kizen is model agnostic. It works with over 50 models, and the list keeps growing, so you can use the right model for each agent and switch as better ones launch, without rebuilding your workflows.
Will Kizen work with our existing systems?
Yes. Kizen plugs into the systems you already run, from your CRM and core platforms to document stores and data warehouses. If there's no ready-made connector, the Integration Agent writes a code step against the system's API so your workflow still reaches it.
What's the difference between an agent and an app on Kizen?
Agents do the work: they classify, extract, research and act inside your workflows, with human approvals where you want them. Apps are the tools your people and customers use, like portals, CRMs and dashboards. Both share the same data, permissions and audit trail, so your agents can run right inside the apps you build.
How long until it's live?
Most teams leave their working session with a working agent or app. From there, a first workflow typically goes live in days to a few weeks, depending on integrations and review requirements, and our team stays with you through launch.
Your first agent could be live this week.
Build it with a Kizen engineer in a Free AI Working Session, or start free on your own.
