The semantic layer between your existing systems and your users, agents and dashboards. Everything audit-logged. Runs in your environment, on-premise or in our secure data centre in Sweden. The diagram below shows the architectural pattern using a pulp/paper example; the same pattern applies to packaging, chemistry, food, pharma and other regulated process industries.
Your users sit at the top, operators, quality managers, planners, AI agents, auditors. AEL Studio sits in the middle, AI mesh, ontology runtime, MCP tools, audit pipeline. Your existing systems sit at the bottom, historian, LIMS, MES, ERP, identity. Everything within a single EU sovereignty boundary that you control.
Example shown: pulp/paper manufacturing. Object types in the ontology runtime (Production_Run, Quality_Test, Deviation, Work_Order, Shift) are domain-specific. Replace these with packaging-, chemistry- or pharma-specific objects and the architectural pattern is identical.
The three-layer split is not arbitrary. It reflects a fundamental separation of concerns: users and their interfaces (top), the semantic and reasoning layer (middle), and your existing systems of record (bottom). Everything within an EU sovereignty boundary you control.
Operators, quality managers, planners, AI agents, auditors. Each gets a purpose-built view onto the same underlying domain, not five separate integrations, one ontology with five lenses.
The semantic layer where data, AI agents and audit fuse into operational action. The ontology runtime defines and enforces your domain model in real time. MCP tools auto-generated from the schema. AI mesh orchestrates agents across model providers.
Process historian, lab data, MES/SCADA, ERP, identity. AEL Studio observes them by default; writes back via typed, audited actions only where you authorise. We don't replace your existing stack, we make it queryable as a coherent whole.
Every component in the middle layer has a specific job. None of them lock you in, most are open-source under permissive licenses, and the AEL-authored runtime is delivered to customers under MIT license.
Defines and enforces your domain model. Schema described in LinkML 1.7 (a published standard). Hot-reload of additive changes, new fields appear in agents, dashboards and APIs within ~1.5 seconds without restart. Typed actions with role-based ACL and audit-by-default. GraphQL and MCP endpoints auto-generated from the schema.
Model-agnostic agent orchestration. Anthropic, OpenAI, Mistral, local models, your fine-tunes, switchable without rewriting agents. Audit and observability follow the agent across provider switches. Agents can call other agents as tools via Model Context Protocol.
For each ontology object: get_, list_, search_. For each action: typed write endpoint with parameters, ACL check and audit emit. The agent's available tools update automatically when the schema changes. No manual prompt engineering for new fields.
Every action is logged twice, to ClickHouse for queryable audit and to Langfuse for LLM trace correlation. Even rejected actions are logged (security with traceability). Auditor can query "who did what, when, on which data, with which model" in one place. CSRD- and EU AI Act-ready.
Object instances in PostgreSQL HA, typed, indexed, transactional. Vector embeddings in Qdrant for RAG over deviations and historical context. RustFS as S3-compatible object storage. All open-source, all running in your environment.
For actions that need approval flows, retry semantics, or undo/compensation logic. Initial deployments often skip Temporal, actions run synchronously through the runtime. Add it when complex workflows justify the operational overhead.
Postgres, Qdrant, RustFS, ClickHouse, Langfuse, OpenTelemetry, LibreChat, Flowise, n8n, Temporal, all open-source under permissive licenses. The AEL-authored runtime (ontology service, MCP builders, schema loader, action runtime, audit pipeline) is delivered to customers under MIT license. AEL retains copyright on the runtime; customers receive permanent usage, modification and distribution rights.
We can walk your IT architects, security team or compliance organisation through the architecture in detail, including deployment topology for your specific environment, integration patterns for your existing systems, and the customer-pull update model that keeps you in control.