Agents Overview

Agents are purpose-built AI assistants configured for a specific business job, not general conversation.

Instead of giving the model a new context every time, an agent is pre-configured with:

Specific instructions: How it should behave, what it should do, and what it must avoid
Trusted data: Your knowledge base, documents, and internal references
Connected tools: Actions it can execute (search, scrape, crawl, system actions, and integrations)

This combination makes agents consistent, reliable, and operational for real work.

Quick Definitions

Agents: Purpose-built AI assistants configured with instructions, access rules, and optional capabilities for a specific business job.
Knowledge Bases: The memory-backed collections and files your agent can retrieve from to ground answers in your organization data. Learn more in Memory System.
Tools & Connectors: Action capabilities that let agents perform operations and integrate with external systems. Learn more in Tool Hub.

What Makes an Agent Different from General Chat

A general chat model is broad and flexible, but it has no built-in understanding of your team policies, workflows, or business context unless you provide it every time.

An Arkios agent, by contrast, is configured once and reused across your team with the same behavior and grounding.

General Chat

Good for open-ended brainstorming
No persistent business role by default
No guaranteed access to internal knowledge
No controlled tool permissions out of the box

Agent in Arkios

Defined role with stable instructions
Grounded in your approved knowledge sources
Can take actions through selected tools
Team-ready controls for visibility and access

How Agents Leverage AI with Your Instructions and Data

An effective agent follows a three-layer pattern:

Instruction Layer
- Defines role, tone, constraints, and output format
- Example: "You are a procurement assistant. Return a short summary, key risks, and recommended next step."
Knowledge Layer
- Provides organization-specific facts
- Example: onboarding docs, SOPs, policy manuals, product guides
Action Layer (Tools)
- Lets the agent retrieve live information or perform operations
- Example: crawl a website, scrape a page, run supported tool actions

When all three are combined, the agent does not just answer questions. It can complete useful tasks.

Why Teams Prefer Agents Over General Chat

Teams usually choose agents when they need consistency, speed, and control.

Consistency: same prompt standards and behavior for every user
Accuracy: responses grounded in selected data sources
Governance: scoped visibility and tool access rules
Efficiency: less repeated prompting, faster outcomes, fewer manual steps

In short, an agent turns ad hoc chat into a repeatable workflow.

Example Use Cases

1. Internal Policy Assistant (Knowledge Base Driven)

Goal: Help employees get policy answers quickly.

Instructions: answer only from policy docs and cite source section
Data: HR policy handbook, compliance wiki, leave policy PDFs
Tools: optional web search disabled to avoid unapproved external content

Result: Employees get fast, policy-aligned answers without searching multiple documents.

2. Sales Research Assistant (Knowledge + Web Tools)

Goal: Prepare account briefs before customer meetings.

Instructions: produce account summary, recent events, risks, and outreach angle
Data: internal playbooks, ICP definitions, previous account notes
Tools: web crawl and web scrape for current company information

Result: Reps spend less time researching and more time on high-value conversations.

3. Support Triage Agent (Knowledge + Operational Tools)

Goal: Classify incoming issues and recommend next actions.

Instructions: identify severity, impacted area, and immediate workaround
Data: product docs, known-issues KB, runbook articles
Tools: retrieval and selected operational tools for diagnostics

Result: Faster triage, better handoffs, and reduced back-and-forth across teams.

Practical Prompt Pattern for Business Agents

Use a structured system prompt template like this:

You are the [ROLE] for [TEAM/COMPANY].
Primary objective: [MAIN GOAL].
Only use approved sources: [KNOWLEDGE/TOOLS].
Response format:
1) Summary
2) Recommended actions
3) Risks/unknowns
If information is insufficient, ask one clarifying question.
Never invent facts. State uncertainty clearly.

Getting Started

Create the agent and define a clear role.
Add a focused system prompt with boundaries.
Connect relevant knowledge and tools.
Test with real prompts from your team.
Publish and monitor usage in Analytics.

Continue with: