01
Identity and policy
Every person, service, and agent acts inside an organization with explicit roles, scopes, policies, and approval authority.
- Organizations and workspaces
- RBAC and attribute rules
- Service identities
- Approval authority
Platform architecture
The operating layer beneath every module and agent.
Runlium is designed as a modular business platform with shared identity, authorization, data, events, search, automation, and AI controls—not a collection of products connected after the fact.
Runlium Core
CRM
Projects
Finance
People
Support
Docs
Four platform layers
The shared foundation creates the compounding value.
Each layer is independently testable and observable, while the user experience remains one coherent product.
02
Business data graph
Common business entities share stable identifiers and relationships instead of being copied into disconnected application silos.
- Organizations and people
- Work and commitments
- Money and documents
- Events and relationships
03
Workflow and integration
Typed events, APIs, connectors, approvals, retries, and durable execution move work reliably between modules and external systems.
- Events and webhooks
- Workflow orchestration
- Integration contracts
- Execution history
04
AI and agent runtime
A provider-independent AI layer retrieves approved context, calls scoped tools, observes limits, and records every meaningful action.
- Model routing
- Retrieval and citations
- Tool permissions
- Evals and cost controls
The business graph
Relationships are first-class product data.
A customer is not only a CRM record. It may connect to contacts, proposals, projects, invoices, support conversations, documents, risks, and renewal commitments. Preserving these relationships makes cross-functional questions possible without exporting data into a separate reporting project.
Stable entitiesShared IDs and explicit source-of-truth rules.
Traceable relationshipsUnderstand how a conclusion connects back to operational records.
Permission filteringGraph traversal never grants visibility beyond the requesting identity.
Northstar Ltd.
CRMRenewal · Oct 18
ProjectImplementation
Finance$12.4k due
SupportPriority case
ActionAgentManagerAdmin
Read account historyAllowAllowAllow
Draft external emailAllowAllowAllow
Send external emailReviewAllowAllow
Issue a refundDenyReviewAllow
Delete customer dataDenyDenyReview
AI governance
Permission-aware answers. Policy-aware actions.
The AI layer should not receive a master credential. It should operate through the identity, tenant, approved tools, budgets, and action policies relevant to the current request.
- Read and write scopes are separate.
- High-impact actions can require a named approver.
- Model, prompt, tool, cost, and result belong in the audit trail.
- Evaluations protect behavior as models and prompts change.
Deployment path
Launch on managed infrastructure. Preserve portability from day one.
The first release should minimize operational burden without tying product architecture to one hosting provider.
Managed cloud
Fastest path for early customers, continuous delivery, managed database, object storage, queues, and observability.
Dedicated environments
Separate database, storage, keys, network boundaries, and deployment controls for larger or regulated customers.
Customer-controlled options
Bring-your-own-cloud or regional deployments after the product, support model, and upgrade process are proven.
Platform-first product