Ofelia for IT
Multi-agent AI, a deterministic BPA engine, and native Slack & Teams integration — in one governedarchitecture. 15+ years of enterprise heritage. Full visibility before you deploy.
This is your entry point to the technical foundations.
Ofelia Agentic
“From docs to workflows.”
Bonita BPM
“Design. Deploy. Scale”
Connect to your ecosystem
One suite, two products, multiple integration layers. From chat adapters to enterprise system execution.
Native Teams and Slack integration. Hexagonal architecture. New channel = new adapter, zero impact on existing services.
Ticketing, email, calendar, and more. User's OAuth token. The AI prepares, the user confirms, the BPA engine executes. Deterministic. Fully traced.
Keycloak SSO. OIDC, SAML, Entra ID, Google. Auto-provisioned on first interaction. One identity, one audit trail.
Configurable connectors and REST API extensions in BPA Studio. Same governance as the conversational path.
Connects to what you already run
LDAP/AD
CRM Platforms
Databases
UIPath
REST APIs
ERP Systems
LDAP/AD
CRM Platforms
Databases
UIPath
REST APIs
ERP Systems
LDAP/AD
CRM Platforms
Databases
UIPath
REST APIs
ERP Systems
LDAP/AD
CRM Platforms
Databases
UIPath
REST APIs
Jira
LDAP/AD
CRM Platforms
Databases
UIPath
REST APIs
ERP Systems
LDAP/AD
CRM Platforms
Databases
UIPath
REST APIs
ERP Systems
BPA Engine. Deterministic by design
Sequencing, assignment, and enforcement are handled by the BPA engine. A BPA Engine with 15+ years of proven reliability in Finance, Insurance, and regulated operations.
Powered by Bonita. 15+ years in production. BPMN 2.0. Every step, assignment, and deadline enforced deterministically. No AI variability in the execution path. Whether triggered from Slack or a dedicated process application, the same engine runs underneath.
Runtime assignee resolution from the live org chart. Per-step timeout, reminders, escalation. Full SLA tracking. Every action traced.
Automatic retry with exponential backoff. Recovery mechanism re-executes stalled elements without manual intervention. Clustered architecture for high availability. Configurable performance tuning to match your workload.
The BPA engine powers conversational execution and orchestration inside Slack and Teams with Ofelia Agentic, and dedicated process applications with Bonita BPM. Same engine. Same governance. Same audit trail.
Multi-agent AI. Specialized, grounded and deterministic
Ofelia's AI agents read your documentation. They answer questions, guide your teams, and suggest workflows. They never act on their own. When execution starts, the BPA Engine takes over — deterministic, traceable, under your control.
Ofelia Agentic doesn’t rely on a monolithic prompt. It runs specialized AI agents orchestrated by a routing graph. The AI layer runs independently in Python (LangGraph + LangChain), upgradeable without affecting other services.
Five agents, one job each.
Every request is routed to a specialist — not handled by a single monolithic prompt. Intent classification, action preparation, knowledge retrieval, scope control, and feedback formatting each run in their own agent. The routing graph decides who handles what. The architecture ensures no single point of AI failure.
Grounded knowledge retrieval.
Vector search, knowledge graph traversal, and key-value lookup, simultaneously. Multi-hop retrieval answers complex queries that pure vector search misses. PDF, DOCX, Markdown, HTML,parsed and indexed automatically. No pipeline to build.
From AI to deterministic execution.
Ofelia Agentic hands off workflow orchestration to the BPA engine. From that point, execution is fully deterministic: sequenced, assigned, and enforced by the process runtime. The AI initiates. The engine guarantees.
Human-in-the-loop.
The AI proposes. The user confirms. For structured workflows, the BPA engine enforces. No autonomous execution: whether the step is AI-guided or engine-orchestrated, the user stays in control.
“Every response grounded in your documents. Every action confirmed by the user. Every step traced. No autonomous execution.”
Enterprise-grade governance.
Built in, not bolted on
From authentication to audit trails, every layer designed for organizations that can't afford to compromise on control.
Identity and access management
Keycloak SSO (Entra ID, Google, SAML, LDAP) with MFA. Dual authorization: RBAC + ABAC. This supports fine-grained policies like 'only Finance department users can view expense reports' without creating a role for every team. Auto-provisioning on first interaction. Per-service data isolation.
Audit and compliance
Full audit trail on processes, actions, AI interactions, and permissions. Process instance audit trail: every step, every actor, every timestamp. Action execution audit: every integration call logged with parameters and results. Soft-delete preserves user identities for post-departure compliance.
Architectural security
API proxy pattern — browsers never see backend tokens or service URLs. Signed pagination cursors (HMAC-SHA256) prevent data enumeration. Structured LLM output prevents injection-style AI vulnerabilities. Event-driven decoupling reduces direct API surface area. CSRF protection enabled by default. XSS input sanitization via OWASP Java HTML Sanitizer. Java Security Policy (ProGrade) protects against malicious script execution.
Observability
OpenTelemetry distributed tracing across all services. LGTM stack — Grafana, Loki, Tempo, Prometheus — deployable with a single command. End-to-end request tracing: from browser click to LLM response and back. LangSmith integration for AI-specific tracing — prompts, token usage, agent routing decisions. Runtime monitoring via Prometheus endpoint.
Built on 15+ years of enterprise reliability
Complete governance, auditability, and scalability for organizations that can't compromise on process execution. Every step logged, every decision traceable, every workflow optimized for performance at scale.