Architecture

An overview of how Arkios Enterprise is structured from agents to teams and infrastructure.

Core Building Blocks

Arkios Enterprise is built around a few core platform layers:

• Agents: Purpose-built AI workers with instructions, memory access, and optional tools
• Memory: Persistent knowledge collections used for grounded retrieval
• Tools: Action capabilities and integrations that agents can invoke
• Teams & Access: Role-based controls for visibility, governance, and collaboration
• Artifacts: Saveable outputs generated from useful responses

Logical Flow

A typical request follows this path:

1. A user submits a request to agent.
2. The selected agent applies its system instructions.
3. If configured, the agent retrieves context from memory collections.
4. If needed, the agent executes tools to gather live data or perform actions.
5. The model composes a grounded response.
6. Useful outputs can be saved as artifacts for later use.

Runtime Layers

1) Experience Layer

This is the user-facing layer where conversations happen.

• Home chat for fast AI interaction
• Agent chat for purpose-specific workflows
• Group chat for team collaboration with optional in-thread AI

See Chat Interface.

2) Intelligence Layer

This layer defines how an agent thinks and responds.

• System prompt and behavior constraints
• Model selection and response settings
• Retrieval and tool-use policies

See Agents Overview and Create Agents.

3) Knowledge Layer

This layer grounds responses in approved business context.

• Memory collections store indexed files and references
• Retrieval fetches relevant chunks at query time
• Answers become more consistent and less generic

See Memory System and Knowledge Sources.

4) Action Layer

This layer enables external operations and integrations.

• Tool definitions and auth strategy in Tool Hub
• Agent-level tool attachment and access controls
• Observable tool-call execution during runtime

See Tool Hub and Tools & Connectors.

5) Governance Layer

This layer controls who can do what, where, and with which resources.

• Team membership and role permissions
• Visibility settings for agents and resources
• Usage and credits management

Architecture Principles

• Separation of concerns: Instructions, knowledge, and actions are configured independently
• Grounded responses: Memory-backed retrieval improves factual reliability
• Controlled execution: Tool access is explicit and governed
• Reuse over repetition: Artifacts and reusable agents reduce duplicate work
• Team-ready by default: Permissions and visibility are built into daily workflows

Next Step

Continue to Quick Start to build your first end-to-end workflow.