- Manual version
- 1.27.0
- Applies to Klarun release
- 2026.09
- Last updated
- October 21, 2026
Documentation
Everything you need to set up, configure, and use Klarun, your data team's operating system.
This documentation covers all six modules (Reporting, Governance, Platform, People, Projects, and FinOps), the 360 Assessment, the Business View, Organization Settings, and the Roles system. Use the sidebar to jump to any section.
What is Klarun
Klarun is a Data Ecosystem OS, a single platform that gives data leaders and their teams full operational visibility over every layer of their data organization: the reports they produce, the assets they govern, the people they manage, the platform they run, and the investments they make.
It is designed for CTOs, heads of data, and BI managers who need to operate at scale without the overhead of custom tooling. Rather than spreading context across Notion pages, spreadsheets, and disconnected BI portals, Klarun centralizes it, and connects to the tools your team already uses (Fabric / Power BI, Databricks and Microsoft 365 Copilot) to sync their metadata and usage automatically.
Klarun follows a hybrid input model: some data is imported automatically through integrations, while other data is registered manually by your team. Both types live in the same interface and can be enriched, tagged, and acted upon in the same way.
Module Summary
Klarun is organized into six modules. Each module covers a distinct area of your data operation.
Starting with Klarun
Follow these steps when setting up Klarun for the first time. The order matters, because organization settings and integrations unlock data that later steps depend on.
Go to Settings → Organization. Fill in your organization name, industry, company size, fiscal year start month, and timezone. The fiscal year drives FinOps, the timezone drives incident metrics, and the industry gives the 360 Assessment context.
Still in Settings → Organization, register your company email domain. Once verified, anyone with a matching email can join via an invite link, with no one-off codes required.
Go to Settings → Membership and generate an invite link. Choose multi-use (whole team), email, domain, or single-use types. Assign roles before or after members join. Everyone joins on a viewer seat, which every plan includes without limit; move the people who need module roles to a data-team seat, which is what your plan counts.
In Settings → Integrations, connect Fabric / Power BI, Databricks, Microsoft 365 Copilot and Azure Cost Management. Once connected, Klarun syncs reports and assets from Fabric / Power BI, assets from Databricks, licences and usage from Copilot, and monthly Azure cost into FinOps automatically. Integrations require the Business plan or higher. To use Klarun from Claude, ChatGPT or Copilot, set up the AI connector (MCP) on the same page; see AI Connector.
In Reporting → Report Catalog, your synced reports will appear immediately. Enrich them with Domain, Subdomain, Owner, and Schedule. Manually register any reports from other tools.
Before inviting business users, hide sensitive workspaces (HR, Finance, Compensation...) in Settings → Report Catalog and choose who still sees them. Everything else in the catalog stays open to everyone.
In People → Data Team, add every member of your data organization. Complete profiles (role, squad, seniority, skills, and languages) to unlock People Ops workforce analytics.
In People → Onboarding, upload documents, paste links, or write text notes in each of the eight onboarding sections. New joiners access everything from this page, and the items business readers need can be published to the Business View's Guides.
In Platform → Architecture, document your data layers and infrastructure components, and expand the Business Domains section at the top of that page to map your domains. In Platform → Health, add any ongoing incidents.
Assets are synced from integrations. In Governance → Contracts & Sources, add your data source contracts and SLAs.
In FinOps → Budget & Plan, enter your budget per platform and category. In FinOps → Actual Cost, record actual monthly spending. Cost Intelligence then computes budget vs. actuals automatically.
In Projects → Objectives, create your strategic goals. Link Key Results to each objective, then create Initiatives (the concrete work items) on the Initiatives kanban.
Free Plan
Klarun offers a free plan that lets you get started without a credit card. It is designed for small teams exploring the platform: all six modules are available, with data entered by hand rather than synced from an integration.
- 2 data-team seats
- Unlimited viewer seats for the Business View
- Modules: Reporting, Governance, Platform, People, Projects, FinOps
Reporting
The Reporting module is the central registry for all reports, dashboards, and data products in your organization. It gives every stakeholder a single place to find what exists, who owns it, how often it refreshes, and where to open it.
Pages: Report Catalog
Reports enter the catalog in two ways: automatic sync from the Fabric / Power BI integration, or manual registration for any other tool. Databricks adds assets to Governance, not reports to the catalog.
Report Catalog fields
Governance
The Governance module provides a unified view of your data ecosystem's health: all platform assets, how they are used, and the contracts governing your data sources.
Pages: Observability Hub · Assets · Contracts & Sources
Assets are populated automatically from the Fabric / Power BI and Databricks integrations, with no manual entry needed. They are enriched with health flags (broken, orphan, stale) computed from integration metadata. Because assets come from integrations, a plan without them shows the Observability Hub and Assets pages with a note on how to get them. After a downgrade, syncing pauses and the synced assets stay readable.
Contracts & Sources are entered manually. Each contract represents a data source your organization depends on. Fill in the fields below to track obligations and risk.
Contract fields
On Corporate and Enterprise, the Fabric tab of the Observability Hub shows how much of each Fabric capacity is used. Klarun reads the Microsoft Fabric Capacity Metrics app's semantic model nightly, with the Azure cost sync: each night it reads the last 14 days of CU seconds and throttled minutes per capacity per day, and the 50 items that used the most CU each day, for up to 10 capacities (any more wait for the next night). Klarun keeps 90 days.
It is set up on the Azure Cost Management connection, with the same app registration: see Organization Settings → Integrations.
The Databricks tab of the Observability Hub shows DBU use per workload (jobs, pipelines, SQL warehouses, clusters and notebooks) over the last 30 days, read from system.billing at list price. These figures are for information only and never become FinOps actuals.
Unused jobs are jobs that spent DBUs writing tables that nothing read in the same window, judged from system.access.table_lineage. Reads that go straight to the storage path, or through a Fabric shortcut or mirror, are not seen there, so check with the table's consumers before stopping a job.
Both need a SQL warehouse on the connection and the system-table grants listed under Organization Settings → Integrations; the panel names the grant that is missing. Asset lineage comes from the same lineage table, over the last 90 days.
Platform
The Platform module is where you document the infrastructure your data team runs on and track its operational health. It covers architecture documentation, business domains, and platform health.
Pages: Architecture · Health
Architecture is an inventory, not a diagram. It is divided into fixed sections, one per layer of the stack, and you register each tool, system, or workload you run as an entry inside the section it belongs to.
Architecture sections
Every entry captures more than a tool name. Fields are grouped so that each workload records who is accountable for it, how heavily it is used, what it costs, and how it is secured.
Business Domains are registered in the Business Domains section at the top of the Architecture page. Tick the domains your data is organized around, or add your own. Domains are referenced across Reporting, Platform incidents, People, and Projects, and the Business View groups reports, incidents, and initiatives by them. A domain written on a report that matches no registered domain, by name or by id, is not treated as one: the Business View still shows it as the report's area, but no incident or initiative is matched to it.
Health incidents are logged with a title, severity (P1 to P4), a category (availability, performance, freshness, data quality, access or cost), the affected service, description, and current status: Open, Investigating, Monitoring (fixed, watching) or Resolved. Update the status as the incident progresses; the first move out of Open is recorded as the acknowledgement time, which is what mean time to acknowledge is measured from. The affected service is chosen from the Architecture registry rather than typed, so renaming an entry keeps its incident history and two entries that share a name stay distinct. Picking one suggests a severity from that service's business criticality, which you can override. Every change of status, severity, assignee or service is appended to the incident's history, which is written by the platform and cannot be edited. Marking an incident Resolved requires resolution notes and a resolved time that is not earlier than the start time, because those two fields are what MTTR is calculated from. A P1 or P2 additionally requires a root cause category and either a preventive action or an explicit "no preventive action needed": those are the incidents most worth learning from and the ones most likely to be closed in a hurry.
The Service map tab, beside Sections on the Architecture page, draws your entries from their Depends On field: each arrow points from a dependency to the entry that uses it. Entries with no dependency either way are not drawn. Switch between the Diagram and Table views.
Select an entry to see its blast radius: every entry that depends on it, directly or through others. The Health column comes from open incidents: a P1 or P2 shows Down, a P3 or P4 shows Degraded, and entries that depend on one show Impacted.
A dependency that would close a loop is refused when saved, and if a loop exists anyway, its entries are tagged Loop. The map shows only what your team has entered: nothing is discovered from your cloud.
People
The People module is the hub for your data organization's human layer. It combines a team registry, workforce analytics through People Ops, and an onboarding resource library.
Pages: People Ops · Data Team · Onboarding
Data Team profile fields
People holds no salary or compensation. Pay is recorded only in FinOps, under Actual Cost.
Onboarding sections
The Onboarding page is organized into eight fixed sections. Admins populate each section with documents, links, or text content.
Resource types
Publishing to the Business View
A People admin can publish any item, in any section, to the Business View's Guides: use Publish to Business View on its card, or the Show in the Business View switch in its form. Publishing asks first, since everyone who can open the Business View will read it. Business readers see the title, the description, and the content, never the purpose, tools, or roles. Published items carry a Published tag and are listed under Published to Business View; Remove from Business View takes one off at once. People admins here are people_admin, org_admin, and org_member, whatever their other roles. Business readers see published items once Guides is switched on in Settings → Business View; items can be published before that.
An admin can narrow a member's People role to Themselves and their reports, in Settings → Membership under the People roles. The member then reads only their own team member record and everyone under it in the reporting line, directly or indirectly. Data Team shows how many of the whole team they are seeing, and the org chart starts at them.
The line starts from the team member record linked to the member's Klarun account. With no linked record, with two, or when the reporting line loops back to them, the member sees nobody, and Settings shows a warning next to them. Managers are checked on every change, so a new loop cannot be saved. The change applies on the member's next request and is recorded in the audit log. A member with org_admin is never narrowed.
Projects
The Projects module is Klarun's OKR (Objectives and Key Results) framework. It gives data teams a structured way to define their strategy, track measurable outcomes, and manage the concrete work items that deliver on those outcomes. The Portfolio gives a high-level view across all objectives and initiatives.
Pages: Portfolio · Objectives · Key Results · Initiatives
Start top-down: create Objectives first, then add Key Results to each one, and finally create Initiatives under Key Results.
An admin can narrow a member's Projects role to Only what they own, in Settings → Membership under the Projects roles. Reading does not change: the member still sees every objective, key result, and initiative, so the Portfolio and the progress roll-ups stay whole. What narrows is what they can change.
An initiative's contributions move the progress of the key results it links. So a narrowed member cannot link, unlink, or change a contribution to a key result they do not own, even on an initiative they own. Those links stay as they are.
Ownership comes from the team member record linked to the member's Klarun account, and each account can be linked to one record only. With no linked record, or with two, the member can change nothing, and Settings shows a warning next to them. The change applies on the member's next request and is recorded in the audit log. A member with org_admin is never narrowed.
FinOps
The FinOps module gives data leadership the financial visibility they need to run the data function as a business. It covers budget planning, actual cost tracking, and a Cost Intelligence overview that compares the two.
Pages: Cost Intelligence · Budget & Plan · Actual Cost
Budget & Plan: Add budget lines for each data platform (e.g. Databricks, Snowflake, Power BI) broken down by cost category and month. The fiscal year calendar is driven by Settings → Organization → Fiscal year start.
Actual Cost: Enter or import actual monthly spending for each platform and category. On Corporate plans and above, AI assist can parse uploaded invoices and payroll into structured monthly rows.
Cost Intelligence reads from your budget and actuals and computes variance automatically. No separate data entry is required on this page.
With Azure Cost Management connected (see Organization Settings → Integrations), the nightly sync writes billed Azure cost into the Platform actuals on FinOps → Actual Cost: one synced line per cost category per fiscal year, marked Synced from Azure. Mapping rules decide which category each cost row lands in.
Matching ignores case. The most specific rule wins, then the lower priority. Cost no rule matches is listed under Unmapped spend, largest first, and Map on a row opens a rule prefilled from it. Adding, changing or deleting a rule rewrites the synced lines at once.
Rules belong to the Platform section: members who hold it see them, and finops_admin changes them. Databricks DBU figures in Governance are list price for information and are never written here.
FinOps data is split into three sections, each shared all-or-nothing on top of the FinOps role. The role sets how much a member can do in FinOps; the sections set which parts of it they see.
With nothing chosen, FinOps admins hold every section and everyone below admin holds Platform and Projects, so payroll is always an explicit grant. A member with org_admin always holds every section. An admin changes a member's sections in Settings → Membership, under the FinOps roles. The change applies on the member's next request and is recorded in the audit log.
Figures that add sections together, such as the combined budget and variance on Cost Intelligence, appear only for members who hold both Platform and Team. Opening or deleting a fiscal year and changing the fiscal-year start need every section. The 360 Assessment shows its finance section only to members who hold every section.
360 Assessment
The 360 Assessment is Klarun's AI-powered organizational review. It synthesizes data from across your modules (Governance, Platform, People, Projects, and FinOps) into a single, periodic assessment of your data organization's health, with narrative insights and recommendations.
Pages: 360° View
There is no manual data entry. Keep the other modules up to date, then generate an assessment for a period. The quality of the result depends on how complete your Governance, Platform, People, Projects, and FinOps data is.
Business View
The Business View is one front door for people outside the data team. It answers four questions: where is the report I need, can I trust it right now, who do I ask, and what is changing that affects me. Everything it shows comes from records your data team already keeps in Klarun, through a fixed list of fields and in plain language: no priorities, incident numbers, platform ids, or other internal terms.
Pages: Home · My App · Find · Status · What's coming · Guides
Apart from My App, which is each reader's own list, every reader gets the same answer: nothing on these pages depends on who is asking. Opening a report and searching add one to the company's counts for the week (reports opened, opened after a search, and searches). In My App, a save that adds reports adds one, a save that removes reports adds one, and each report shown adds one, however many reports a save changes. No reader, id, report, or search word is stored with them, and nothing a business reader does changes any data but their own My App list.
A reader adds a Power BI report with Add to My App on its card in Find, up to 50 reports, and orders them as they like. Each report opens inside Klarun in Power BI's own frame: the reader signs in to Power BI there and sees only what their own Power BI access allows. Klarun stores no Power BI token. With Power BI sign-in for My App on in Settings → Business View, Add to My App first asks Power BI, with the reader's own sign-in in their browser, whether they can open the report: a reader without access is told so instead of the report being added, and new access can take a few minutes to apply. With it off, or when the check cannot run, the report is added unchecked, and Power BI says so when a report the reader cannot open loads.
My App takes Power BI and Fabric reports only. Paginated reports, dashboards, apps, and reports in other tools say Not available for My App and still open in their tool. The frame needs Embed content in apps allowed in your Power BI tenant; if it is not, the frame stays blank and Open in Power BI still works. The frame's address carries the report's Power BI ids, which anyone who can already see the report in Find can read in the browser's developer tools. Nobody else sees a reader's list, the data team included, and it is deleted when the reader leaves the company.
Each report in Find carries at most one badge: the first of these that applies. No signal means no badge, never a claim that all is well.
Incident badges appear only while Status is on. Endorsements share one row on the card: Certified in green (badge 4, so only while no issue badge shows), Promoted in blue, and Built on master data in gold, when the report's semantic model is endorsed as Master data in Fabric. Only data items can be master data, never a report.
While a published initiative of type Report Migration, Tech Migration, or Data Migration is in progress or planned, every report in its area says that some reports there are moving, with the planned completion month, and links to the initiative. The notice speaks of the area, never of the report, and appears only while What's coming is on. Area badges and notices need a Business Domain: an area that is not one gets neither.
Everything is set in Settings → Business View. Members with org_admin or org_member manage it, and they are the only ones who can open Settings. Every change applies to everyone at once and is recorded as admin.business_view_changed, naming what changed but never the contact.
In Projects → Initiatives, use Publish to Business View on an initiative's row, or the Show in the Business View switch in its form. Anyone who can edit an initiative can publish it, so a member limited to what they own publishes only their own. Publishing asks first, because a title business readers have seen cannot be taken back. Published initiatives carry a Published tag, and the Published to Business View filter lists them. Delivered work leaves What's coming 90 days after its end date.
In People → Onboarding, a People admin publishes items one by one, in any of the eight sections, with Publish to Business View or the Show in the Business View switch. It works like initiatives: whoever manages Settings switches Guides on or off for the whole company (it starts off), and People admins choose what goes on it. Publishing asks first, a Published tag marks what business readers see, and Remove from Business View takes an item off at once. Guides shows only the sections that hold a published item, and an empty page while nothing is published. See People.
A report's area is its Business Domain from Platform → Architecture when its domain matches one, by id or by name ignoring case and spacing. Otherwise it is the report's own domain value, because the domains in Fabric need not match the ones registered in Klarun; spellings that differ only in case or spacing are one area. Incidents and initiatives are matched to Business Domains only, so an area that is not one gets no incident badge for the area and no migration notice. A domain that is empty or reads as an id shows no area.
Organization Settings
Settings are accessible from the sidebar under your organization name. The page is divided into five tabs: Organization, Integrations, Membership, Business View, and Billing. Users with the org_admin role can modify every setting. The org_member role can manage the Organization, Integrations, Membership, and Business View tabs, but not Billing, and cannot grant the admin role.
Organization tab
Configure your organization profile. These values appear throughout the app: the fiscal year drives FinOps, the timezone drives incident metrics, and the industry gives the 360 Assessment context.
Domain verification
Register your company email domain to enable domain-based invite codes. Anyone with a matching email address can then join your workspace using a domain invite link, without needing a one-off code.
Verification is done via a one-time code sent to an admin email on the registered domain. Once verified, the domain is locked and used for all future domain invites.
Integrations tab
Connect your BI tools to Klarun. Synced integrations automatically populate Reports (Reporting module), Assets (Governance module) and the Observability Hub. Members with org_admin or org_member connect, sync, and disconnect them; module roles, reporting_admin included, do not. Connecting and syncing need the Business plan or higher. Disconnecting works on any plan: after a downgrade, syncing pauses and what was synced stays readable.
Power BI setup steps
Go to Azure Portal → Microsoft Entra ID → App registrations → New registration. Give it any name (e.g. "Klarun Integration"), select Single tenant, and click Register. Use a dedicated registration: do not reuse the one connected to Microsoft 365 Copilot.
Leave API permissions empty. Do not add Tenant.Read.All, and do not grant admin consent. Power BI refuses a service principal that carries admin-consented Power BI permissions in Entra, and the call then fails with a 401 that reads like a credential problem. Access is granted in the Power BI admin portal instead, in steps 03 and 04.
Open Power BI Admin Portal → Tenant settings → Developer settings. Enable "Service principals can call Fabric public APIs". This covers workspace and report listings. If you scope it to a security group, add the service principal to that group.
Power BI Admin Portal → Tenant settings → Developer settings → Service principals can call Fabric public APIsIn the same portal, open Tenant settings → Admin API settings and enable "Service principals can access read-only admin APIs". This is a separate switch from step 03 and covers tenant-wide metadata, lineage and usage. Scope it to a group the service principal belongs to, then allow about 15 minutes for both settings to take effect.
Power BI Admin Portal → Tenant settings → Admin API settings → Service principals can access read-only admin APIsBack in the app registration → Certificates & secrets → New client secret. Set an expiry and click Add, then copy the Value column immediately. Do not copy the Secret ID, which sits beside it and is not a credential. The Value is shown only once.
Go to Settings → Integrations → Power BI → Connect. Enter your Tenant ID, Application (Client) ID, and the Client Secret value. Click Test connection, and if successful, click Save & sync.
Databricks
Connect one or more Databricks workspaces with the workspace URL and a service principal or a personal access token, and optionally a SQL warehouse. Once connected, Klarun syncs catalogs, schemas, tables and views, clusters, SQL warehouses, jobs, pipelines and notebooks as Governance assets. Usage and activity come from the workspace audit log (the system.access.audit system table), which Klarun can only query through a SQL warehouse: when you choose one, usage syncs too; without one, the assets sync and usage does not.
DBU usage, unused jobs and lineage read Unity Catalog system tables through the same SQL warehouse, filtered to the workspace. Grant the service principal USE CATALOG on the system catalog, and USE SCHEMA and SELECT on system.billing (usage and list prices) and system.access (audit and table lineage). USE SCHEMA and SELECT on system.lakeflow are optional and add job run failures to the Health suggestions. This needs a Unity Catalog workspace, and a workspace URL of the adb-NNNN form, which carries the workspace id; otherwise these reads are skipped.
Microsoft 365 Copilot
Use a dedicated Entra app registration, not the one connected to Power BI: Copilot's admin-consented Graph permissions would stop a shared registration from reaching Power BI's read-only admin APIs. In the registration, under API permissions, add the Microsoft Graph application permissions Reports.Read.All and Organization.Read.All, then grant admin consent. Settings can copy the tenant ID from your Power BI connection.
Klarun reads licence counts and tenant-wide Copilot usage reports. Reports.Read.All covers every Microsoft 365 usage report, since Microsoft offers no Copilot-only scope. Microsoft publishes usage with a 48 to 72 hour delay, so the first figures appear a few days after connecting.
Azure Cost Management setup steps
Go to Azure Portal → Microsoft Entra ID → App registrations → New registration. Give it a name (e.g. "Klarun Cost"), select Single tenant, and click Register. Keep it separate from the Power BI and Copilot registrations. It needs no API permissions.
On each subscription Klarun should read, open Access control (IAM) → Add role assignment, choose Cost Management Reader, and assign it to the app. The role is read-only: it shows cost data and grants no access to the resources themselves. A new assignment can take a few minutes to apply.
Subscription → Access control (IAM) → Add role assignment → Cost Management ReaderIn the app registration → Certificates & secrets → New client secret. Set an expiry, copy the Value (not the Secret ID) right away, and note the expiry date: you will need it in step 06.
Go to Settings → Integrations → Azure Cost Management → Connect. Enter the Tenant ID (a GUID; common and organizations are refused), the Application (Client) ID, and the Client Secret value. Test connection lists the subscriptions the app can read.
Active subscriptions are preselected. Keep up to 50 and click Save; a subscription the app cannot read is refused. Cost is read nightly, per month, resource group and service. Azure publishes cost 8 to 24 hours late, so the latest days fill in over the next syncs.
Klarun does not track the expiry date. Before it, create a new secret in the same registration, then Connect again, test, and save: saving replaces the stored connection. Then delete the old secret in Azure. If the secret expires first, the nightly sync fails, the connection shows the error, and the cost already synced stays.
Membership tab
Manage who belongs to your Klarun workspace. You can view all current members, change their roles, and create invite codes.
Seat types
Every member holds a viewer seat or a data-team seat. Everyone who joins, by invite code or by domain, starts on a viewer seat, which holds only the business_viewer role and opens the Business View. Viewer seats are unlimited on every plan. A data-team seat holds module roles, and data-team seats are what a plan is sized by: 2 on Free, 15 on Business, 30 on Corporate, and unlimited on Enterprise. The Members header shows how many are in use. When all of them are, a member on a viewer seat cannot move to a data-team seat until someone moves back to a viewer seat or the plan changes. Reclaiming admin access through DNS always succeeds, even on a full plan, since it is how an organization with no admin gets one back.
Invite code types
Business View tab
Choose which Business View pages are on, Guides included, and where business readers take their questions. What Guides shows is published item by item in People → Onboarding. The tab also shows how complete the Business View is, how it is used, and a preview, one panel under another. See Business View.
Billing tab
View your current plan and manage your subscription. Upgrading, downgrading, and entering payment details are all handled from this tab. Only users with the org_admin role can access the Billing tab.
Notifications
The bell in the header shows what needs your attention, on every seat. Its count refreshes about once a minute while Klarun is open in a visible tab. Open a notification to go straight to the item; if the item is gone, the page says it is no longer available. An access request can be approved or dismissed from the notification itself. My Account → Notifications lists every notification, with filters, selection, mark as read and delete. A deleted notification can be restored for 10 seconds.
What you are notified about
Each notification goes only to the people who can act on it or read the item, and the list checks this again every time it loads: a member who loses access to an item stops seeing its notification. Nobody is notified of their own action, such as an incident they opened.
How long they stay
A notification stays for 3 days after it last appeared, read or not. An action item (an access request, an incident assigned to you, a catalog review due) stays until someone handles it, then 3 days more. A repeat of the same event updates the existing notification rather than adding another.
Members on a data-team seat can also receive notifications by email, chosen per category in My Account → Notifications. Each email carries the notification's title and a link to the item. Viewer seats never receive email.
The same page can hide alerts and updates from the bell. Critical incidents always appear, and action items and system notices cannot be hidden. The catalog review keeps its own daily reminder email, described in Hidden Catalog Items.
An org admin can turn on Email critical incidents to the whole data team in Settings → Organization. P1 and P2 incidents then reach every data-team member who receives them by email, including those who turned alert emails off or hid the type.
AI Connector
The AI connector lets Claude, ChatGPT, Microsoft Copilot and other assistants that support the Model Context Protocol (MCP) read Klarun and, when your company allows it, add and update items in it. Each person connects with their own Klarun account, and the assistant sees only what that person's roles allow. The connector URL is in Settings → Integrations → AI connector (MCP). The connector is available on every paid plan; on the Free plan an org admin can still turn it off and revoke clients, and members can still disconnect their apps.
Company access
An org admin chooses what assistants may do for the whole company, in Company access on the same panel. It starts at Read only.
Connect an assistant
Each person adds Klarun in their assistant with the connector URL, signs in to Klarun and approves the access the assistant asks for. Claude and ChatGPT need no setup in Klarun.
What assistants add
A report an assistant adds lands as Pending Review, so a reporting admin checks it before it counts. Anything an assistant adds or edits shows an Added via or Edited via tag (Claude, ChatGPT or Copilot) in Report Catalog, Budget & Plan and Actual Cost, and Report Catalog can filter by who added a report. Fields a sync fills, such as a synced report's details or the lines the Azure cost sync owns, are not written over. An assistant sends at most 50 items per call, and sending the same item again does not create a duplicate.
Disconnect and review
Anyone can disconnect their own assistants in My Account → Connected AI apps. An org admin sees every connection under Connections on the AI connector panel and can revoke any of them, or revoke an OAuth client and every connection made through it. Download AI connector activity (CSV) lists each tool an assistant called, by whom and through which client, with the number of items; what was asked is never recorded. Setting Company access to Off disconnects everyone at once.
Roles
Klarun uses a layered role system. Every member has one org-wide role that controls baseline access across the entire workspace, plus optional module-specific roles that grant elevated permissions within individual modules.
Org-wide roles
Assigned at the workspace level. Apply to all modules unless overridden by a module-specific role.
Module-specific roles
Each module has three tiers: admin, contributor, and viewer. Module roles are additive: assigning a module role to a member grants that permission level for that module regardless of their org-wide role.
Permission matrix
Baseline permissions by role tier, applicable across all modules. Module admins have the same privileges as org admins within their module scope. Exception: in People and FinOps, creating, editing, and deleting records needs the admin tier, and FinOps is further divided into sections (see FinOps sections). In Projects, an admin can limit a member to changing only what they own (see Edit only what you own), and in People to themselves and their reports (see Themselves and their reports). Anyone whose Klarun account is linked to a team member record can edit their own skills and languages. The last four rows are organization settings: they belong to org_admin and org_member only, never to a module admin.
Data Access Scopes
Roles decide what a member can do in each module. Data access scopes narrow which of that module's records they reach. A scope only ever narrows a role: it never grants access the role does not give, and it never applies to a member with org_admin. Scopes belong to each member, are set by an admin in Settings → Membership next to the roles they narrow, and are enforced on the server on every request, for the pages and the API alike. Report Catalog items are narrowed per item rather than per member: see Hidden Catalog Items.
The three scopes
Each scope is described in full with its module: see FinOps sections, Edit only what you own, and Themselves and their reports.
When a scope cannot be worked out
The Projects and People scopes start from the team member record linked to the member's Klarun account. Each account can be linked to one record only, and an admin can link a record to their own account only when the record's e-mail is theirs. When the link is missing or doubled, or when the reporting line loops back to the member, the scope gives nothing rather than guessing: in Projects the member can change nothing, and in People they see nobody. Settings shows a warning next to the member so an admin can fix the record. Managers are checked on every change, so a new loop cannot be saved. A stored scope that cannot be read is treated as the narrow choice.
Everywhere the data appears
A scope follows the data rather than the page, so the same rules apply wherever that data is shown.
Audit
Every scope change is recorded as admin.scope_changed, naming the member, the admin, and the new value for each module it touched. Every change to the account a team member record is linked to is recorded as people.link_changed, and role changes remain admin.role_changed. A scope change applies on the member's next request without signing them out; a role change signs them out of every session, so it applies at once.
These events sit in the organization's audit log, which records sign-ins, sign-outs, and failed sign-ins, plus security and admin changes: roles, data access scopes, members removed, sessions signed out, integrations connected, disconnected, and synced, contract changes, Business View settings, the Copilot seat model and per-user Copilot exports, the AI connector (its access level, registered clients, and connections made or revoked), team members editing or linking their own record, and decisions on proposed initiatives (ROI): approvals, deferrals, rejections and re-baselines, without the amount or the rationale. It records those events, not every change to every record. Members with org_admin or org_member, who manage Settings, can download it as CSV from Settings. Each tool an AI assistant calls through the connector is recorded too, never what was asked, and downloads on its own from the AI connector panel in Settings, Integrations.
Hidden Catalog Items
The Report Catalog is open to everyone in your organization by default. An org admin or a reporting admin can hide a Power BI workspace or a single report in Settings → Report Catalog, or from the row menu in the catalog, and choose the people who still see it. Admins always see everything. Hiding is a Klarun setting: it does not read or change permissions in Power BI or Microsoft Fabric.
How hiding works
Where it applies
A hidden item is left out of every list, search, count, usage ranking and export for people not on its list: the Report Catalog, Business View, Governance assets, health, activity and owners, the asset CSV export, and the owned assets on a People profile. Opening its link directly gives the same answer as an item that does not exist. In lineage it appears as Unavailable item, with no name, workspace or owner. People on the list see a lock badge naming who else sees it. Jobs that run without a signed-in person, such as 360 Assessment and incident suggestions, leave out every item under a rule. If the rules cannot be read, Klarun hides every ruled item from everyone except admins rather than showing them.
Keeping lists current
When a member is removed from the organization, or their People record is marked inactive or removed, they leave every list at once. Every 90 days each list is due for review: Settings marks it, an admin confirms it in one click, and org admins and reporting admins receive a daily reminder email with the number of lists due and a link, never the item names. Each rule keeps its full change history, which admins can view in Settings or download as CSV.
Customer-deployed Edition
In the customer-deployed edition Klarun runs in your own Azure subscription, as one installation for your one organization. Your data, logs and backups stay in your tenant and region, and the installation sends nothing to Klarun: no telemetry and no heartbeat. Klarun and your infra team install and upgrade it together, following the install and upgrade runbooks sent with each release. This section is for your Klarun administrator and your Azure team.
What is different
Prerequisites
Versions and upgrades
Releases are named by date (CalVer, YYYY.M.PATCH, for example 2026.10.0). Your address followed by /api/healthz shows the running version, and the version block at the top of this manual shows the release it describes. Each release comes with notes in English, Portuguese and Spanish: the image digest, the database changes, whether rolling back is safe, and what changed. Upgrades go one release at a time, take about 30 minutes, and apply database changes before the new version takes traffic. Klarun supports the current release and the two before it (N-2).
Rollback and restore
If a new release misbehaves, your team moves back to the previous release by its image digest. Database changes only add, so a rollback needs no restore. Azure keeps 35 days of database restore points in your region. A restore goes into a new server and the installation is then redeployed onto it, so the live database is never overwritten. Run a restore drill after install and once a year.
Support
Your infra team is the first line. When something fails they run the support bundle script, which reads versions, health, job history and recent logs with read-only access, and replaces tokens, keys, email addresses and IP addresses. It never reads secrets, settings or database rows. Your team reviews the archive and sends it through the support channel in your contract.