Enterprise AI guardrails, governed and sovereign.
Every AI request runs through guardrail policies you can read, test and audit. And the whole platform can run where your rules require: your public cloud account or your own data centre.
Two promises
Control what AI does. Control where it runs.
Enterprise AI needs two kinds of trust. Trust in behaviour: it follows your rules. Trust in location: your data stays where you decide.
Governed
Guardrails on every request
Policies cover access, sources, consistency, grounding, output quality and content safety, and each one is visible and auditable.
18 built-in guardrail policies
Human approval for sensitive actions
Full activity log for people and agents
Sovereign
Your environment, your rules
Deploy Kroolo in your own AWS, Google Cloud or Azure account, or on premises, with the same governance in every case.
Public cloud or on-premises
Data stays in the region you choose
Keys, network and access under your control
Guardrail policies
Six pillars of trustworthy output.
Guardrails are not a filter added at the end. They run through the full path of a request, and each policy has an ID, a severity and a clear rule you can inspect.
GR1
Access Control
Know who is asking, always.
The requesting user is resolved from the active session before anything is retrieved. Answers never reach wider than that person's own permissions.
GR2
Data Provenance
Every figure has a source.
Metrics and data points carry the connector, dataset and timestamp they came from. If the origin cannot be traced, it is not shown.
GR3 · GR4
Consistency
One number, everywhere.
A value is computed once and reused by every agent, chart and export, and narratives must match the numbers exactly.
GR5 · GR6
Hallucination Prevention
Say less than you know, never more.
Trends need enough data. Root-cause claims must cite the records behind them. Below the threshold, the agent refuses or flags low confidence.
GR7 · GR8 · GR9
Output Integrity and Transparency
Show the uncertainty and the freshness.
Forecasts show confidence bounds. KPIs carry full metadata. Every widget shows when it was last refreshed.
GR10
Content Safety
Block what should never be generated.
Explicit, violent, hateful or illegal content is stopped at ingestion and at generation, and recorded.
Policy library
Read every rule, before it runs.
Each policy is listed with its category, severity and exactly what it enforces. Security, risk and compliance teams can review the full library, not a black box.
Clear IDs and severity
Critical, High and Medium ratings show what matters most.
Search and filter
Find policies by pillar or severity, and see which are active.
Linked to activity
Every blocked or flagged output points back to the policy behind it.
Decision path
Every request, five checkpoints.
The same path applies to a person in chat and to an agent running overnight. Nothing is shown, sent or changed without passing it.
1
Identity
The requesting user is resolved from the session.
GR1
2
Retrieve
Only permitted sources are searched, with provenance kept.
GR2
3
Generate
Grounding, consistency and sufficiency checks apply.
GR3 to GR6
4
Check output
Confidence, metadata, freshness and content safety.
GR7 to GR10
5
Record
The decision and its policy are written to the activity log.
Audit
Allow
The output is rendered and logged.
Flag or ask
The output is marked low confidence, or a named person approves it.
Block
The output is withheld and the person sees which policy applied.
Activity and audit
Every decision leaves a record.
The Activity view lists each request with who asked, what happened and which policy decided it. Records cover people and agents, and they are built for review.
Who and what
The person or agent, the request and the time.
Which policy applied
The rule that allowed, flagged or blocked the step.
Sources and model
What was read, and which model handled it.
Exportable
Filter and export for security and compliance reviews.
Sovereign deployment
Run it where your rules say.
Kroolo can be deployed on a public cloud or on premises, so data residency, security and regulatory requirements decide where AI runs, not the vendor.
Public cloud
Your account, your region
AWSGoogle CloudMicrosoft Azure
Deploy into your own AWS, Google Cloud or Azure account
Choose the region that meets your residency rules
Use your cloud security, networking and monitoring tools
On-premises
Your data centre, your hardware
Your data centrePrivate network
Run inside your own facilities and network
Keep data and AI processing within your perimeter
Operate under your existing security and change processes
Your people
Web, mobile, API
Your environmentAWS · Google Cloud · Azure · On-premises
Conversation
Agents
Trust
Knowledge
Connectors
Guardrail policies, approvals and audit log run inside the same boundary
Your tools
Slack, Drive, Jira
Deployment governance
Same governance, wherever it runs.
Choosing where Kroolo runs does not change how it is governed. The policies, approvals and audit trail travel with the deployment.
Data residency
Choose where data and AI processing live, down to the region or facility.
Your keys
Encrypt with keys you own and rotate, so access can be revoked on your side.
Network isolation
Keep traffic on private networks and inside the boundary you define.
Approved models only
Policies decide which models may handle which data, in every deployment.
Autonomy
Trust is earned, one step at a time.
Start with suggestions and give agents more room as results prove out. You set the level per agent, team and workflow.
Suggest
The agent proposes and a person decides. Best for new workflows and sensitive data.
Act with approval
The agent prepares the action and waits for a named approver before it runs.
Act within policy
Low-risk, well-understood work runs on its own, inside limits you set and with a full record.
Mapped to the frameworks your auditors use
Our Responsible AI framework sets out nine principles, risk tiers and lifecycle controls, and maps them to the standards below.
Read the Responsible AI frameworkNIST AI RMFISO/IEC 42001EU AI ActGDPRSOC 2 / ISO 27001
FAQs on AI Governance
See governed AI in your environment.
Bring your requirements. We will show the guardrails, the audit trail and the deployment options that fit.
Explore
More from Kroolo AI
- LLM Routing and AI Cost Control by TeamKroolo routes each AI request to a fast, balanced or frontier model, fails over when one is slow, and shows AI spend by team, workflow and agent. See it live.
- Enterprise AI StackSee how Kroolo's enterprise AI stack works: connectors, a knowledge graph with vector and keyword search, permission checks, agents and cited answers.
- Agentic AI Workflows, Governed and TracedCompose agents, skills, tools and human approvals on one canvas. Test with evaluations, ship with permissions and audit on, and trace every run.
- Context StudioDefine, test and govern the definitions, policies, playbooks and sources Kroolo AI uses. Versioned, permission-aware and shared by every agent.
- Enterprise AI SearchAsk one question, get a cited answer from 50+ apps like Slack, Drive, Jira and Salesforce, limited to what each person can see. SOC 2. Book a demo.
- Multi-Model AI PlatformUse Claude, GPT and Gemini in one governed workspace. Kroolo matches each task to the right model, controls access by team and supports your own API keys.