Conversation Context
Maintains the conversational state required for the current interaction.
AgentOS platform
The enterprise runtime behind every OniLeo agent.
AgentOS is the architecture that connects customer channels, agent execution, enterprise knowledge, business systems, AI services and enterprise control into one coordinated platform.
AgentOS architecture
One architecture from conversation to enterprise execution.
OniLeo separates customer channels, agent execution, AI services and enterprise systems into coordinated platform layers. Business logic remains within the agent runtime instead of being tied to a single channel or AI provider.
Layer 1 · Experience Channels
Web ChatBrowser VoiceSIP / Phone Voice
Customer interaction surfaces send conversational events into the same agent runtime.
Layer 2 · Agent Runtime
Intent & Routing
Determines where the current customer input belongs in the active agent journey.
Workflow State
Maintains the active workflow and current progression.
Information Capture
Collects required business information during structured journeys.
Decision Logic
Evaluates configured decision paths.
Knowledge Routing
Coordinates enterprise knowledge requests with the active journey.
Action Execution
Invokes controlled business actions and API operations.
Handoff Coordination
Routes conversations to human assistance when required.
Response Coordination
Produces the appropriate next conversational response while preserving runtime state.
Layer 3 · Enterprise Knowledge
- Approved content
- Semantic retrieval
- Relevant business context
Layer 3 · AI Services
LLMEmbeddingsSTTTTS
Layer 4 · Enterprise Systems
APIsCRMTicketingDatabasesBusiness Applications
Connected through a controlled action and integration boundary.
Stateful orchestration
Keep the active customer journey intact.
Enterprise conversations are not a sequence of isolated prompts. OniLeo maintains the active conversational and workflow state so each customer turn can be interpreted in the context of the journey already in progress.
Customer Input↓Conversation Context↓Active State
IntentWorkflowKnowledge
↓Next Runtime Action↓ResponseContext preserved across turns
- Customer input is evaluated against the current context.
- Active workflow state remains available while the runtime handles relevant knowledge requests.
- Collected information remains associated with the active journey.
- Decisions and actions are evaluated in the correct workflow context.
- Runtime can resume the appropriate next step after an interruption or supported detour.
Knowledge orchestration
Knowledge is part of the runtime, not a separate chatbot.
OniLeo coordinates enterprise knowledge retrieval with active workflow state. This allows the agent to answer relevant business questions while retaining the customer journey that was already in progress.
- Active workflow: waiting for preferred appointment date
- Customer asks: what time does the clinic close?
- Knowledge Routing to Enterprise Knowledge, then contextual answer
- Resume active workflow: continue with preferred appointment date
Workflow execution
Execute structured business journeys through a common runtime.
Generated workflows are executed by the AgentOS runtime as stateful customer journeys. The runtime coordinates information requirements, decisions, knowledge steps, actions and handoffs according to the active workflow configuration.
Information
Collect required customer and business data.
Decision
Evaluate configured branches.
Knowledge
Retrieve approved enterprise information when required.
API Action
Invoke an approved external operation.
Message
Communicate the appropriate next step.
Handoff
Transfer to human assistance when required.
Workflows executed here originate from the agent configuration and generated workflow design.
Enterprise execution
Connect agent decisions to business systems.
The AgentOS action layer provides the boundary between conversational workflow execution and enterprise applications. It allows configured agent journeys to retrieve business information, trigger actions and update external systems through controlled integrations.
Agent Runtime↓Action Execution↓Controlled API Layer↓
APIsCRMTicketingDatabasesBusiness Apps
Retrieve · Validate · Trigger · Update · Complete
Human assistance
Escalate without discarding the conversation.
When a workflow or customer situation requires human assistance, OniLeo can coordinate handoff as part of the active agent journey and preserve relevant conversational context for the transition.
AI Agent Runtime→Handoff Decision→Context Package→Human Assistance
Conversation Context · Captured Information · Active Journey · Handoff Reason
Multilingual runtime
Language can change. The journey should not.
OniLeo separates conversational language from the underlying business journey so agents can support multilingual interactions while retaining the relevant workflow and conversational context.
Customer speaks in their language↓Language Understanding / Speech Layer↓Shared Agent Runtime↓Workflow · Knowledge · Actions↓Agent responds in the appropriate conversation language
- Language handling does not require duplicating the business workflow for every language.
- Voice and text language processing connect to the same underlying runtime.
- Context remains associated with the active journey when conversational language changes.
Channel architecture
Separate the customer channel from the agent logic.
Channel adapters connect web and voice experiences to the same AgentOS runtime, allowing business workflows and enterprise knowledge to remain independent of the customer interaction surface.
Web Chat Adapter
Browser Voice Adapter
SIP / Phone Adapter
↓
AgentOS Runtime
Speech Input → STT → Agent Runtime → TTS → Speech Output
AI provider layer
Keep business logic separate from AI-provider choice.
OniLeo provides a common intelligence layer for allocating language, semantic and speech capabilities without embedding the enterprise workflow directly into one model or provider.
Reasoning & Generation
Used where generative reasoning or response generation is required.
Semantic Understanding & Retrieval
Supports semantic matching and enterprise knowledge retrieval.
Speech Recognition
Converts supported voice interactions into runtime input.
Voice Generation
Produces conversational speech output for supported voice channels.
Provider-flexible. Workflow-stable.
Enterprise control plane
Separate agent execution from platform control.
The OniLeo control plane manages the configuration and operational controls surrounding agent execution, while the runtime handles live customer interactions.
Enterprise Configuration
Enterprise-level platform settings.
Agent Configuration
Configuration associated with each agent.
Version Management
Maintain controlled agent configuration versions.
Publishing Control
Control which approved version is available to production channels.
Provider Configuration
Manage the AI service configuration available to the platform.
Usage & Cost Visibility
Track AI and provider consumption where available.
Operational Traceability
Make runtime activity available for investigation and improvement.
Runtime plane
- Conversation Processing
- Workflow Execution
- Knowledge Retrieval
- Decision Execution
- Actions
- Handoffs
- Responses
Enterprise foundation
Operate multiple agents without mixing their business context.
OniLeo is designed around an enterprise environment where multiple AI agents can serve different business purposes while retaining their own configuration, knowledge allocation, workflows, channels and publishing state.
Enterprise
Agent AAgent BAgent C
Configuration · Workflows · Knowledge Allocation · Channels · Version · Runtime State
Enterprise separation
Keep enterprise configuration logically scoped.
Agent isolation
Maintain agent-specific purpose and runtime configuration.
Controlled publishing
Release approved agent versions deliberately.
Shared platform services
Use common platform capabilities without duplicating the underlying AgentOS foundation.
Controlled releases
Keep production separate from work in progress.
Agent configuration changes should move through a controlled version and publishing boundary so teams can refine an agent without silently changing the experience already running in production.
Working Version→Validation Boundary→Approved Version→Published Runtime
Testing is performed through the OniLeo product tooling. This section explains the platform release boundary, not the Test Studio interface.
Runtime telemetry
Make agent execution traceable.
AgentOS produces operational signals across the customer interaction and runtime path so teams can understand how an interaction was routed, which workflow executed, whether knowledge or actions were used, and how the response was produced.
Customer Input→Intent / Route→Workflow State→Knowledge / Decision / Action→Response
Interaction Events
Customer and agent interaction events.
Workflow Events
Workflow progression and outcomes.
Knowledge Events
Knowledge retrieval activity and misses where captured.
Action Events
Configured business action execution.
Handoff Events
Escalation activity.
Latency Signals
Runtime and service timing information where measured.
AI Usage
Provider and model consumption and cost information where available.
Detailed dashboards live in Bot Canvas, Agent Operations.
One AgentOS foundation
One runtime. Multiple intelligence services. Connected enterprise execution.
OniLeo brings stateful orchestration, workflow execution, enterprise knowledge, business actions, multilingual processing, channel adapters, AI services and enterprise controls together through a common AgentOS architecture.