Data Academy · Layer 11 · Reason & act

Agent Fabric

Build, deploy and orchestrate domain agents.

Lesson 13 of 17 · Find it on the platform map

01 What it is

What this layer does

The agent fabric is where domain agents are built, registered, run and coordinated. An agent is software that takes a goal, plans steps, calls tools and works with people to complete a task. SCIKIQ provides an agent builder, a registry of domain agents, a runtime with tracing, orchestration for multi-agent systems, a tool registry and human-in-the-loop controls, with templates for sales, finance, operations, supply chain, customer service and risk.

02 Concepts

Four ideas to hold on to

Agent registry

A catalogue of every agent, what it is for, which domain it serves and what tools it may use. SCIKIQ keeps an agent registry of domain agents, supported by agent cards for agent-to-agent interoperability with remote agents.

Tools and planning

Agents act by calling tools, such as APIs or other services, chosen as part of a plan to reach the goal. SCIKIQ maintains a tool registry covering HTTP, OpenAPI, MCP and plugins.

Orchestration

Coordinating several agents, or several steps of one agent, so work is divided and results combined in a controlled way. SCIKIQ supports orchestration and multi-agent systems, with streaming and asynchronous runs for longer tasks.

Human-in-the-loop

Designed points where an agent stops for a person to supply information or approve a step before continuing. SCIKIQ lets an agent pause, ask, wait for approval and resume.

03 In SCIKIQ

What the layer contains

Agent builderAgent registry and domain agentsAgent runtime with tracingOrchestration and multi-agent systemsShort- and long-term memoryTools and planning: tool registry (HTTP, OpenAPI, MCP, plugins)Agent templates: Sales, Finance, Operations, Supply chain, Customer service, RiskStreaming and asynchronous runsHuman-in-the-loop: pause, ask, approve, resumeAgent-to-agent interoperability (agent cards, remote agents)

04 What good looks like

Signs it is working

Every agent in use appears in the registry with a named domain, purpose and list of permitted tools.
Any agent run can be replayed step by step from its trace, including tool calls and decisions.
Sensitive steps pause for a named human approver before the agent continues.
New agents start from a template or shared components rather than being written from scratch.

05 Diagnostics

Questions to ask your team

1
How many agents are running in our organisation today, and where is that list kept?
2
Which tools can each agent call, and who approved that access?
3
At which steps must a person approve before an agent proceeds, and is that enforced or just intended?
4
If an agent made a mistake yesterday, how quickly could we trace exactly what it did and why?

06 Keep going

Frequently asked

What is the difference between an agent and an automation workflow?

A workflow follows fixed steps defined in advance; an agent is given a goal and plans its own steps, choosing which tools to call. Agents still benefit from fixed guardrails and human approval points.

Why does an agent need both short- and long-term memory?

Short-term memory holds the state of the current task; long-term memory carries useful knowledge across runs. Together they stop agents repeating questions or losing track of earlier work.

Can agents from other platforms work with SCIKIQ agents?

The fabric supports agent-to-agent interoperability through agent cards and remote agents, so agents can discover and call one another across boundaries.