Book a demo Sign up

The enterprise AI stack: five layers, one intelligence.

Kroolo AI is built as five layers that work together. Connectors bring in your data, Knowledge makes sense of it, Trust controls access, Agents do the work, and Conversation is where you ask. Every layer is explained below.

Layer 5Conversation
Ask in plain language, get a cited answer or a finished action.
Layer 4Agents
Role-based agents plan the work and use tools to do it.
Layer 3Trust
Permissions, policies and audit applied to every request.
Layer 2Knowledge
One index of everything your company knows, with its context.
Layer 1Connectors
Live connections to the tools where your work already lives.
Your tools and data+ 40 more
How it works

One question, five layers.

Follow a single request down through the stack and back up. Each layer does one job, and the answer is only as good as the layer beneath it.

1
Conversation
You ask: "What is blocking the Q4 launch?"
Summarise open risks this week
Kroolo AI
2 risks are high: vendor delay and budget.
2
Agents
A delivery agent plans the steps and picks its tools.
Review Q3 launch planDone
Sign off budgetIn review
Brief the design teamTo do
3
Trust
Only what you may see is allowed through. The request is logged.
Access, enforced at retrieval
Finance teamRead
Sales leadsRead, edit
ContractorsNo access
4
Knowledge
Graph, vector and keyword search find the best passages.
Summarise open risks this week
Kroolo AI
2 risks are high: vendor delay and budget.
5
Connectors
Live data is read from Jira, Slack and Drive.
Connected sources5 live
Synced 2 min ago
The answer travels back up the same path: cited sources, then an agent action if you ask for one, all inside your permissions.
1Layer 1 · Connectors

Connect every source.

Everything starts with your data. Connectors link Kroolo to the tools your teams already use, so nothing has to be copied or migrated.

In plain words
Connectors are the plugs. They reach into Slack, Drive, Jira and the rest, and bring what they find into Kroolo.
Content and context. Messages, files, tickets and records, with who wrote them and when.
Permissions travel with it. Each item keeps the access rules from its source tool.
Always current. Changes sync continuously, so answers reflect today, not last month.
Why it matters. Without connectors, AI only knows what you paste into it. With them, it knows how your company actually works.
Connectors50+ native
SlackSynced
Google DriveSynced
JiraSynced
GitHubSynced
SalesforceSynced
NotionSynced
ConfluenceSynced
GmailSynced
Brings in
Content
Keeps
Permissions
Stays
Up to date
Knowledge
Q4 launch blockers
Knowledge graph
Who, what, how related
Vector search
Meaning, not keywords
Keyword (BM25)
Exact names and IDs
Merged and ranked
Q4 launch planConfluence0.94
LAUNCH-212: payments reviewJira0.91
#launch-war-room threadSlack0.88
2Layer 2 · Knowledge

Make sense of it all.

Raw data is not knowledge. This layer organises everything connectors bring in, so the right passage can be found fast and for the right reason.

In plain words
Knowledge is the memory. It links people, projects, documents and conversations into one searchable picture.
Knowledge graph. Maps how people, projects and documents relate.
Vector search. Finds passages by meaning, even when the words differ.
Keyword search. Catches exact names, IDs and phrases. All three are merged and ranked.
Why it matters. Better retrieval means better answers. The model gets the right context, and every claim can point to its source.
3Layer 3 · Trust

Permissions before answers.

Trust sits between what Kroolo knows and what anyone can see. Every request is checked before any content reaches a model or a person.

In plain words
Trust is the gatekeeper. If you could not open it in the original tool, Kroolo AI will not show it to you either.
Permission-aware. Access is checked first, so private content never enters an answer.
Guardrail policies. Rules decide what agents may do, and what needs a human to approve.
Audit trail. Requests, sources and actions are recorded and reviewable.
Why it matters. This is what makes AI safe to switch on for the whole company, not only a pilot group.
TrustPolicy on
Priya asks a question4 sources match. Permissions are checked first.
Q4 launch planConfluenceAllowed
LAUNCH-212: payments reviewJiraAllowed
Board financialsDriveNot shared with you
HR: compensation reviewDriveNot shared with you
Request logged for audit
External sends need approval
4Layer 4 · Agents

Work gets done by agents.

Agents turn answers into outcomes. Each one has a role, a set of tools and a clear task, and works inside the limits Trust sets.

In plain words
Agents are the teammates. You give them a goal and they plan the steps, use your tools and report back.
50+ role-based agents. Ready-made for project, product, sales, marketing, HR and more.
Build your own. Create a custom agent with your context, tools and models.
Human in the loop. Sensitive steps wait for approval before anything is sent or changed.
Why it matters. The same layers that make answers trustworthy make actions trustworthy.
AgentsRunning
Delivery Agent
Task: weekly status for the Q4 launch
ToolsJiraSlackGoogle Drive
Read the sprint board and open issues
Found 3 blockers across 2 teams
Drafted the status update with sources
Waiting for your approval to post
Approve and postEdit draft
Ask KrooloCited
What is blocking the Q4 launch, and who owns each blocker?
Three blockers remain. Payments review is waiting on Security (owner: Dana). The pricing page copy needs sign-off. The data migration depends on the vendor API fix.
Launch planLAUNCH-212#launch-war-room
Create tasksPost update
Ask a follow-up
5Layer 5 · Conversation

Just ask.

The top layer is the one you see. Ask in plain language, in Kroolo Work, Kroolo Search or the mobile app, and everything below works for you.

In plain words
Conversation is the front door. Questions go in, cited answers and finished actions come out.
Natural language. No query syntax or filters. Ask the way you would ask a colleague.
Cited answers. Every answer links back to the sources it came from.
From answer to action. Turn a response into tasks, updates or a drafted message in one click.
Why it matters. People use what is easy. A simple front door is what gets the whole company using the stack.
At a glance

The whole stack, in one view.

LayerIts jobWhat you get
ConversationWhere you askCited answers and actions
AgentsWho does the workPlans, tools and approvals
TrustWhat is allowedPermissions, policy and audit
KnowledgeWhat Kroolo knowsGraph, vector and keyword retrieval
ConnectorsWhere data lives50+ native integrations

FAQs on AI Stack

What is the Kroolo AI stack?
It is the five layers behind Kroolo AI: Connectors, Knowledge, Trust, Agents and Conversation. Each layer does one job, and together they let AI answer and act using your own data.
Do I need to use all five layers?
They work as one system, so you get all five. You choose which connectors to turn on, which agents to use and what policies apply.
Does Kroolo AI show me data I cannot normally access?
No. Permissions from your source tools are applied before any content is retrieved, so answers only draw on what you are already allowed to see.
How is this different from a chatbot?
A chatbot answers from general knowledge. Kroolo AI answers from your connected company data with citations, and its agents can take governed actions in your tools.

Build on the whole stack.

See all five layers working on your own data in a 30-minute walkthrough.

Explore

More from Kroolo AI

Kroolo AI overview