Skip to content
Arketic Agents

Agents: a digital workforce that doesn't just think — it executes

An assistant informs; an agent gets the job done. Arketic agents are triggered by an event, a schedule or a command; they carry multi-step tasks end to end across your ERP and CRM systems, pause for your approval at critical points, and log every step they take.

Human-in-the-loop approval ERP, CRM and ITSM connectors Audit trail on every step
LIVE · Procurement Agent · Task running
Purchase request received Trigger: request form
Stock & budget check in ERP Out of stock · budget OK
Comparing supplier quotes 3 quotes · best offer: Supplier B
4 Manager approval Threshold: the limit you define
5 Create order & report ERP record
Human approval required: amount over the approval threshold — routed to the manager Manager approved — the agent is completing the rest of the task
✓ Task completed · order posted to ERP · audit trail recorded
7/24 Always-on execution
Audit trail recording
Every step logged Approval point Rollback

One thinks, the other executes

The Assistant is an Advisor: Analyzes data, cites sources, and provides decision support.
The Agent is a Digital Worker: Takes ownership of tasks, acts within enterprise systems, and delivers results.

The Thinker

Assistant Decision-support system

  • Runs when you ask; never triggers itself
  • Guides you with cited sources from the corporate knowledge base
  • Summarizes options and, on request, opens a ticket or books a meeting — but doesn't own the process end to end
  • Best at: Q&A, research, summaries, decision support
The Doer

Agent Digital worker

  • Triggered automatically by an event, a schedule or a command
  • Makes decisions, starts the task and follows through to the result
  • Performs real transactions in ERP, CRM and ITSM systems
  • Pauses at critical steps for human approval; logs every action

Not rivals — complements: both run on the same platform, sharing one knowledge base and one permission model. The assistant prepares the recommendation, the agent finishes the job — together they form your digital workforce.

Automation that takes the task, runs it and reports back

Arketic agents are designed for the realities of enterprise processes: they connect to your systems, follow your rules, consult a human at the critical moment and account for every move they make.

Human-in-the-loop

Autonomy doesn't mean losing control. When an amount, risk or data-sensitivity threshold is crossed, the agent stops and asks the authorized person; it won't take the next step without approval.

Approval thresholdsApproval chain

Audit trail

Which agent took which step, when, and with what data — all of it is logged in immutable records. A wrong step is rolled back in one click; your audit team watches everything from a single panel.

Immutable logRollback

Multi-step automation

Dozens of steps from a single instruction: it gathers data, compares, opens records, drafts emails and reports. It carries context between steps and resumes an interrupted task where it left off.

End-to-end tasksContext carry-over

Triggers

The agent doesn't wait for you: an incoming email, a form submission, an ERP event, a scheduled job or a threshold breach sets it off automatically. You can also start it manually.

EventScheduleThresholdCommand

ERP / CRM integration

It doesn't just read — it writes: it opens purchase requests in the ERP, updates opportunities in the CRM, closes tickets in ITSM. Ready-made connectors and REST APIs link it to your existing systems.

SAPSalesforceJiraSlackITSMSQLREST API

RAG-grounded decisions

The agent never decides in a vacuum: with RAG architecture it consults your procedures, contracts and historical records. The document behind each decision shows up in the log.

Corporate memoryCitations
Ready-Made Agents

A ready-made agent gallery for your department

Execution-focused examples from your company's AI universe: each agent ships with its own triggers, system connections and approval rules — then gets tailored to your organization. Setting up a new use case typically takes 1-2 days.

HR

Recruiting & Selection Agent

Scores incoming CVs against the job criteria, runs the pre-screening, sends interview invitations to suitable candidates and hands HR a reasoned shortlist.

CV scoringPre-screeningInterview invites
HR

Leave & Shift Agent

Evaluates leave requests against balances and the team calendar, catches conflicts, updates the shift plan — and escalates exceptions to the manager with the reasoning attached.

Leave processingShift planningConflict checks
Finance

Financial Risk Assessment Agent

Scans customer and supplier data, producing a risk score from payment history and market signals. When a threshold is crossed it alerts the team and drafts the mitigation proposal.

Risk scoringEarly warningPortfolio watch
Finance

Credit & Limit Tracking Agent

Continuously monitors customer credit limits; holds orders at breach risk, updates limits with collection data, and kicks off the limit-increase approval flow when needed.

Limit monitoringBreach blockingApproval flow
Finance

Budget Building & Tracking Agent

Compares spending against budget lines in real time and spots deviations early — then goes to the department manager with the deviation report and a suggested action.

Deviation analysisReal-time alertsSuggested actions
Finance

Automated Reporting Agent

At period close it pulls the data from your systems, prepares the management report and distributes it with an executive summary — then answers follow-up questions from the report itself.

Period-close reportsAuto-distributionExecutive summary
70+ ready-made roles These 6 examples come from the HR and finance universe. The same pattern extends to manufacturing, IT, procurement and sales — for the thinking roles, see the Assistants section below.

Speed comes from autonomy — trust comes from control

Giving an agent the power to act doesn't mean handing over control. In Arketic, every agent operates inside defined boundaries — what it may do, where it must stop and who it has to ask are set from day one.

  • Permission boundaries Each agent's accessible systems, permitted operations and spending limits are defined up front. It can't take a single step outside that scope — the role-based permission model applies to agents too.
  • Approval mechanisms You set thresholds by amount, risk class or data sensitivity; when one is crossed the agent stops and submits to the approver in the chain. The next step won't run until approval arrives.
  • Traceability Which agent made which decision, with what data, based on which document — every step is written to immutable audit records and monitored from a single panel, ready for KVKK and GDPR reporting.
  • Pause, modify, rollback A running task can be paused at any moment, resumed with updated parameters (modify) or handed to a human; wrong steps are rolled back in one click. In an emergency, the whole agent fleet pauses with a single command.

Common questions about Agents

The questions we hear most from teams evaluating agent projects. For anything else, you can talk directly with our engineers in a demo call.

Three safety layers are in place. First, the agent can only act within its defined permission scope — critical operations never run without approval anyway. Second, because every step is written to the audit trail, you see exactly where the error happened within seconds. Third, the faulty step is undone via rollback and the task is handed to a human. So even in the worst case, the damage is limited to a single detectable, reversible step.

No — you design the approval points. Routine, low-risk steps (gathering data, comparing, drafting) flow fully automatically; at critical points such as amount thresholds, sensitive-data access or hard-to-reverse operations, the agent stops and asks. As trust grows, you loosen the thresholds and expand the automation scope gradually.

Ready-made connectors cover ERP, CRM and ITSM platforms — including SAP, Salesforce, Jira and Slack; anything with a SQL database or a REST API can be connected. Agents both read and write in these systems — opening orders, updating records, closing tickets. Every tool call is logged; integrations run inside your network under your permission model, and average go-live time is 48 hours.

If the need is knowledge, start with an assistant; if it's execution, start with an agent. 'Employees should find answers fast' points to an assistant; 'this process should run itself' points to an agent. Most organizations start with an assistant and add agents on the same knowledge base — and in the 4-5 week ROI-focused PoC you test both scenarios on your own data and decide on measured results.

Assistants

The multilingual, voice-and-text face of Agents

Assistants live on the same platform: they take the reasoning power of agents and turn it into a colleague every employee can talk to — by voice or text, in their own language. Below: assistant capabilities, ready-made roles and field results.

A colleague that goes far beyond Q&A

Arketic assistants are built for enterprise reality: they connect to the right knowledge, give the right answer to the right person, and do the work themselves when needed.

Multilingual by default

Natural conversation in 25+ languages, including Turkish. A customer asks in Arabic, the assistant answers in Arabic — and the management report still lands in Turkish.

TRENDEARFR+20

Voice + text

The same assistant both chats and talks. Field workers can't type with gloves on — they ask out loud and hear the answer read back.

Voice modeText modeSpeech-to-text

Corporate knowledge base (RAG)

Connects to your procedures, manuals, contracts and databases. Every answer cites its source — not a hallucination, a document.

SharePointDriveSQLCitations

One assistant, every channel

The same assistant runs on your website, your intranet, Microsoft Teams and WhatsApp. Build the knowledge base once, get consistent answers on every channel.

WebMobileTeamsWhatsApp

Permission awareness

With the Corporate Hierarchy model, the assistant knows who it's talking to: it gives budget details to the CEO but not to the intern — even for the same question.

RBACSSO / SAMLDepartment-scoped access

Taking action

It doesn't stop at answering: it checks calendars and schedules meetings, opens ITSM tickets, and condenses a long thread into two paragraphs. Critical actions still pass through your approval.

Schedule meetingOpen ticketSummarize

Don't build from scratch — start from 70+ ready-made roles

Over 70 ready-made enterprise personas covering every department from HR to Finance — each role ships with its own knowledge-base template, sample tasks and permission profile. Fine-tune it for your organization and go live in 48 hours.

HR Assistants

From Recruiting & Selection to Leave & Shifts, from Performance Review to Employee Onboarding: they take over HR's routine, file leave requests on your behalf and guide new hires.

Recruiting & Selection Leave & Shifts Training & Mentoring

IT Support

Opens tickets and resolves most of them without a human: password and MFA resets, VPN issues, setup guidance. What it can't solve, it escalates to the right team with a summary.

Incident ticket Password reset VPN setup

Sales Assistant

Fluent in your product and pricing; drafts the proposal in minutes, builds competitor comparisons, and updates the CRM for you after every meeting.

Proposal draft Product comparison CRM update

Legal Assistant

Flags risky clauses in contracts, speeds up legislation and case-law research, and answers KVKK questions with cited grounds. Confidential files stay on-premise.

Clause scan Case-law brief KVKK question

Finance Assistants

Cash-Flow Forecasting, Budget Building & Tracking, Automated Reporting, Credit & Limit Monitoring: they report budget lines by permission — the team's right hand at month-end close.

Cash-flow forecast Budget tracking Automated reporting

Customer Service

Answers customer questions 24/7 in their own language: order tracking, returns, product info. Hands complex cases to a live agent — summary included.

Order tracking Returns Live handover

Manufacturing / Field

Explains fault codes, reads maintenance instructions aloud and summarizes shift reports. Field crews reach knowledge by voice — without taking their gloves off.

Fault code Maintenance steps Shift summary

Executive Assistant

Plans your day: sets up meetings, prioritizes and summarizes email traffic, and puts the critical KPIs in front of you in the morning brief.

Schedule meeting Daily digest KPI briefing
Coming soon These are 8 highlights — the catalog holds 70+ ready-made roles across HR, Finance, Manufacturing, IT, Procurement and Sales. Detailed pages for each are on the way.
How It Works

From idea to live assistant in four steps

You start without writing code; every step is managed from a visual interface. Average setup: 48 hours for one department.

1

Connect the knowledge base

Drag in SharePoint, Drive, databases and documents; the assistant learns your corporate memory in minutes.

Docs, SQL, API
2

Define the role and permissions

Start from a ready-made role; tune the tone, the scope and who can access what according to your corporate hierarchy.

Role + RBAC
3

Publish to channels

Go live on your website, portal, Teams and WhatsApp with one click. The same assistant, the same standard on every channel.

Web · Teams · WhatsApp
4

Monitor and improve

Track resolution rate, satisfaction score and missed questions on the KPI dashboard; with feedback, the assistant gets better every week.

KPIs + scores
Your Data Stays Yours. Always. Every assistant inherits the platform's security backbone — and if you choose, all of it runs in your own data center on H200 GPU infrastructure.
KVKK Data localization GDPR Compliance controls EU AI Act Risk classification On-premise H200 GPU infrastructure Department isolation Parallel assistants, zero data bleed Audit logs Every interaction recorded

What assistants deliver in the field

Gains reported from projects in the field, plus the product's measurable facts.

%45

Smaller HR inbox

Routing routine HR questions to assistants cut inbox volume by 45%.

25+ langs

One assistant, every language

International teams and customers get the same standard of service in their own language.

48hrs

From setup to live

With ready-made role templates, the first assistant goes live in 48 hours on average.

7/24

No office hours

Assistants are on duty 24/7 — no question backlog piling up after hours.

Common questions about Assistants

The questions we hear most on assistant projects. For anything else, talk directly to our engineers in a demo call.

The assistant grounds its answers exclusively in your knowledge base and cites the source in every reply. When it can't find a document, it says 'I don't have that information' instead of guessing, and escalates the question to the right team. Low-scoring answers are also flagged on the KPI dashboard — you see the gaps in the knowledge base and close them.

The same assistant runs on your website, your intranet portal, Microsoft Teams, WhatsApp and your call-center line. Behind the scenes it connects to your ERP, CRM, ITSM and HR systems through ready-made connectors, performing actions like opening tickets and running queries directly in those systems. Average integration time is 48 hours.

Yes — and that's by design, not a bug. With the Corporate Hierarchy model, the assistant knows the role and department of whoever it's talking to: the finance director sees budget details, while other employees only get information within their own permission scope. Permissions flow automatically from your existing identity infrastructure via SSO/SAML — no separate lists to maintain.

Both are possible. Ready-made roles (HR, IT Support, Sales, Legal…) come with a knowledge-base template, sample tasks and a permission profile — the fastest way to start. Or you can begin from a blank assistant and define its tone, scope and connections entirely yourself. Most of our customers start from a ready-made role and make it their own over time.

Put your first agent in the field in 4-5 weeks

Pick one process with our ROI-focused PoC model: we design the agent together, run the pilot on your own data, and you decide on measured results.