Bring software, equipment, and signals into the same event contract.
Explore connectorsBuild on Fibric.
Give your operators the context and controls to work across systems.
Early access
Explore the Fibric runtime
OperatorTypeScript
type Operator = (
event: EventEnvelope
) => ExecutionPlan;Context in.A proposed plan out.
Turn system context into a structured execution plan.
Operator referenceExecutionPlan
actions[ ]Policy checkConnector · Action · Limits
ALLOWExecute
BLOCKStop
ALERTReview
Single-flight per entityIdempotency keys
Check permissions and limits before a connector acts.
Execution and trustOne runtime.
Many systems.
Fibric is infrastructure for AI operators across software and equipment. An operator uses connected context to propose a plan; the runtime checks its permissions before execution.
System contextSoftware
Equipment
Signals
Equipment
Signals
Defined responsibilityOperator
Reason with a model.
Propose a structured plan.
Fibric runtimeValidate.
Execute.
Record.
Execute.
Record.
Reference architecture for the generalized kernel. BearScope uses product-specific services.
Explore the architecture
The Fibric kernel
The runtime
behind your
operators.
Build the operator’s role and your product’s experience on a common foundation for context, reasoning, and action.
Explore the kernel architectureReference architecture
Generalized kernel · Managed early access
Applications
Your interface Domain logic Customer experience
Operators
Defined roles Context Proposed plans
Fibric kernel
Shared runtime contract
Event contextSource & tenant identity
Model routingReasoning behind an interface
ExecutionValidated plans & policy
Tenant scopeSingle-flightIdempotencyAction records
System accessCapability interfaces
Software · Equipment · Signals
Runtime backendsReplaceable servicesModels · Execution · Events · Storage
Select a component to explore its contract.
Tenant context throughout.
Carry customer identity from the source event to the proposed action. Data, policy, and credentials belong to the tenant.
Explore tenancyThe model proposes. Code decides.
Validate plans against policy before acting. Single-flight and idempotency controls govern overlapping and repeated work.
Explore execution controlsKeep infrastructure replaceable.
Put models, event transport, storage, and execution services behind defined interfaces. Bind system access by capability.
Explore runtime interfacesFrom context to action.
How it worksReference architecture
SoftwareEquipmentSignals
EventEnvelope
TenantSourceEntity
Keep the source in view.
Every event carries its source, customer, and record or device.
Read the event contractInspect the source event
- Customer
- Tenant identity
- Source
- Connected system
- Entity
- The record or device
- Context
- Source event & related data
Source context
OperatorProposed planCapabilityParametersScope
Give each operator a task.
The operator proposes a plan with the actions and inputs it needs.
Explore operator rolesInspect the proposed plan
- Role
- Defined operator responsibility
- Capability
- Requested operation
- Parameters
- Inputs for that operation
- Scope
- Customer & entity
Check before acting.
The executor checks permissions and coordinates repeated or overlapping work.
Review the control modelInspect the execution controls
- Scope
- Does this belong to the tenant?
- Permission
- Is this capability allowed?
- Coordination
- Is work already in progress?
- Identity
- Is this a repeated request?
Execution record
Request
Decision
Reported result
Keep a record of the result.
Link the request, decision, and reported outcome. Flag unconfirmed results.
Read about execution recordsInspect the execution record
- Request
- Original event & proposed plan
- Decision
- Allowed, deferred, or denied
- Result
- Reported connector outcome
- Reference
- Source & entity identifiers
Actions under your control.
Set the permissions. Fibric checks them before an action runs.
Tenant scope
Keep each customer’s data and permissions separate.
Concurrent work
Coordinate work on the same record or device.
Repeated requests
Recognize repeat requests before running the same work again.
Uncertain outcomes
Hold unconfirmed outcomes for review.