Fabric
The intelligence layer for people.
- Knows the work.
- Knows the context.
- Executes activities.
- Measures productivity.
- Learns continuously.
- Role
- Activities
- Chat
- Skill
- Execution
Onni Enterprise AI Architecture
Your AI strategyneeds an architecture.
Not another stack of experiments.
Move at AI speed. Without leaving your enterprise behind.
Explore the architectureONNI
FabricPeople
BuilderBusiness
Enterprise AI architecture
Two frameworks.One enterprise architecture.
The architecture that lets your AI strategy evolve faster than your technology stack.
AI has to moveat two speeds.
Builder
The transformation layer for the business.
- Understands processes.
- Redesigns operations.
- Builds capabilities.
- Orchestrates agents.
- Puts AI into production.
- Intent
- Canvas
- Process
- Capabilities
- Agents
- Execution
One architecture underneath.
Not two isolated products: two ways to consume the same enterprise foundation.
Your enterprise should ownits AI architecture.
Models change. Vendors change. Agents change. Your architecture shouldn't.
Models
GPTGeminiClaudeLlamaMistral
Cloud
Google CloudAWSAzureOn-premSaaS
Agents
LangGraphADKCustom agentsVendor agents
Onni enterprise AI architecture
Agnostic by design.
- Models are replaceable.
- Agents are discovered.
- Tools are decoupled.
- Systems are encapsulated.
- Knowledge belongs to the company.
Your enterprise already exists.AI has to work with it.
Don't replace everything. Make everything intelligent.
New AI worldAgents · LLMs · MCP · Apps
Onni agent fabricCapabilities · Runtime · Governance
Integration busQueries · Commands · Events
Enterprise reality
- SAP
- Oracle
- Salesforce
- TASY
- Java
- Mainframe
- SEI
- BigQuery
- Data lake
- APIs
- Documents
- SaaS
Anti-corruption layer
- Legacy ERP
- Manifest + driver
- Capability
New connectors plug in through a manifest and a driver. The core never changes.
Integrate what works.Refactor what matters.Build what is missing.
One architecture. From experience to infrastructure.
- ExperienceApps · Chat · Copilots · APIs · Voice
- Agent orchestrationPlanner · Workflows · HITL · Runtime
- CapabilitiesSkills · Tools · MCP · Entitlements
- Context & ontologySources · Context · ACL · Enterprise map
- Integration busEvents · Commands · Queries · Connectors
- Data & knowledgeSQL · Vector · BM25 · Graph · Memory
- EdgeGateway · APIs · Webhooks
- Cloud-native infrastructureKubernetes · SQL · Kafka · CI/CD
- Security · Identity · Governance · Audit
Experience
AI where work happens.
Fabric · Builder · Apps · Chat · Copilots · APIs
Agent orchestration
From intent to coordinated execution.
Planner · Workflows · HITL · Agent runtime · LLM gateway
Capabilities
Separate what the business needs from who executes it.
Skill registry · Tools · MCP · Entitlements
Processes require capabilities. Not hard-coded agents.
Context
Generic intelligence becomes enterprise intelligence.
Sources · Policies · Ontology · ACL · Context
Integration
One controlled path to every enterprise system.
Query · Command · Event
Knowledge
Your enterprise knowledge becomes operational.
Documents · Vector · BM25 · Graph · Memory
Infrastructure
Cloud-native foundation. Enterprise-grade operation.
Kubernetes · PostgreSQL · Kafka · CI/CD · Workload identity
One controlled path betweenAI and the enterprise.
- AI
- Agents
- Skills
- Workflows
- Apps
Integration bus
- ERP
- CRM
- EHR
- Data
- APIs
- SaaS
- Legacy
QueryGoverned read.
CommandDurable action.
EventThe outside world triggers AI.
Skills and screens never call external systems directly. Reads, writes and outside facts all go through the bus.
AI doesn't bypass enterprise architecture.It becomes part of it.
Built for failure.Designed for recovery.
- Action
- Transactional outbox
- Kafka
- Router
- Enterprise system
- Fail
- Retry
- Retry
- Recover
- Fail
- 5× retry
- DLQ
- Reprocess
- Transactional outbox
- Idempotent consumer
- Event streaming
- Retry + backoff
- DLQ
- CQS
- Anti-corruption layer
Innovation speed. Enterprise reliability.
Your knowledge shouldoutlive every model.
- Policies
- Documents
- Procedures
- Data
- Decisions
- Precedents
- Business rules
Enterprise knowledge
- Enterprise
- Department
- Role
- User
- Task
Each level receives only the context it is authorised to see.
Relational source of truth · hybrid vector + BM25 search · graph · memory by scope
Models generate.Enterprise knowledge grounds.
Your AI changes. Your knowledge compounds.
Choose technology.Don't become dependent on it.
Models
GPTGeminiClaudeLlama
←→Model gatewayAgents
LangGraphADKCustom agents
←→Agent fabricTools
MCP serversREST APIsFunctions
←→MCP / APIsSystems
SAPOracleSalesforceTASY
←→Integration busData
PostgreSQLBigQueryVector stores
←→Data layerCloud
Google CloudAWSAzureOn-prem
←→Runtime- Model-agnostic.
- Agent-agnostic.
- Cloud-ready.
- Legacy-compatible.
- Protocol-driven.
Decoupled by design. Today's implementation makes concrete technology choices; each one can be replaced without touching your knowledge, capabilities or policies.
Governance isn't added after AI. It's embedded into execution.
- Identity
- Tenancy
- RLS
- DLP
- Encryption
- Audit
- LGPD
- Policy
- HITL
- Trace
- A CPF arrives
123.456.789-00 - Before the LLM
<PERSON_7F32> - The LLM works
<PERSON_7F32> - Only at the authorised boundary
<PERSON_7F32> → real value
Enterprise AI sovereignty is notabout owning a model.
It's about owning your architecture.
- Your knowledge.
- Your capabilities.
- Your policies.
- Your integrations.
- Your evidence.
- Your decisions.
- Your ability to change.
From the first conversationto the last transaction.
- Experience
- Intent
- Agents
- Capabilities
- Knowledge
- Integration
- Systems
- Data
- Evidence
Complete.
Not “we do everything”: a coherent architecture across the entire AI execution path.
- Build
- Run
- Integrate
- Govern
Two frameworks.One operating model for enterprise AI.
ONNI
FabricPeople AI
- Know the work
- Execute activities
- Ground in context
- Measure productivity
- Learn continuously
BuilderBusiness AI
- Understand the process
- Redesign the operation
- Build capabilities
- Compose agents
- Govern production
Enterprise architecture
Architecture is not the outcome.Velocity is.
- Move fasterFrom idea to production in shorter cycles.
- Keep controlSpeed without losing governance.
- Preserve investmentsLegacy keeps producing value.
- Stay flexibleModels, agents and vendors can change.
- Compound knowledgeEvery execution adds to the company's intellectual capital.
Don't build an AI portfolio.Build an AI capability.
AI projects
- Pilot
- Pilot
- Pilot
- Pilot
- Pilot
- Different stacks
- Different vendors
- Different rules
- Different data
- Different governance
AI architecture
Enterprise AI
Capabilities · Knowledge · Policies
Shared architecture
Fabric
Builder
Every use case
- Your models will change.
- Your agents will change.
- Your systems will change.
- Your ability to change is the architecture.
Move at AI speed.Without leaving your enterprise behind.
Fabric · Builder · Enterprise architecture