---
title: "VOICE — Virtuoso Open Interoperable Context Engine"
description: "VOICE combines Virtuoso, OPAL, and LOAM (Linked Open Agent Memory) into a loosely coupled context engine for AI agents: governed access to existing Data Spaces, persistent memory, and the ability to act, through open standards. Any agent, any LLM."
url: urn:osdi:this
dateModified: 2026-10-30T23:52:06Z
language: en
---
Context Engine · Virtuoso · OPAL · LOAM

# VOICE

Your Data. Your Know‑how. *Any Agent. Any LLM.*

Enterprise AI doesn’t need another data silo. It needs governed access to existing knowledge, persistent memory, and the ability to act.

VOICE (Virtuoso Open Interoperable Context Engine) combines [Virtuoso](urn:osdi:sitevirtuoso/), [OPAL](urn:osdi:siteopal/), and [LOAM (Linked Open Agent Memory)](urn:osdi:siteloam/) into a loosely coupled infrastructure for AI agents. It doesn’t require enterprise data, AI agents, or LLMs to converge on a single proprietary platform: it connects them through open standards, shared semantics, and fine-grained governance.

[See the architecture](#architecture) [Talk to an Architect](urn:osdi:sitecontact/?dept=Solutions%20Architect)

What the animation shows  **Three parts come together.** OPAL modules (agents, skills and tools: tools are exposed over MCP and OpenAPI, skills carry the know-how), agent-rdf-memory (Turtle memory loaded as named graphs, so agent memory becomes data that SPARQL can query) and Virtuoso (a quad store with a SPARQL endpoint, the foundation that holds the graphs and answers queries). **An agent reaches the OPAL modules.** A skill guides it, tools are called over MCP or OpenAPI, and a skill and a tool turn the request into a query aimed at the memory graphs, which are consulted. **The data spaces answer.** Databases are relational tables queried through SQL and RDF Views: rows stay in the database and Virtuoso maps them on demand. APIs are web services called through OpenAPI-described operations, and the response is returned as data the agent can use. **The answer returns the same way:** data space, Virtuoso, OPAL, agent. **One way in.** Databases, knowledge bases, filesystems and APIs share it: knowledge bases (SPARQL endpoints) and filesystems (WebDAV, S3) join the same way.  [Open the full-size animation](http://localhost:8890/skin-openlink-2026/img/native/voice-context-engine-assembly.svg)

Architecture

## Your enterprise already has Data Spaces. Your AI agents should work across them, not require their replacement.

Most AI-ready data platform strategies bring data, analytics, Knowledge Graphs, and AI capabilities closer together within one ecosystem. VOICE starts from a different premise: existing systems stay where they are, and agents reach them through open interfaces.

voice-assembly-animation:start

voice-assembly-animation:end

Federated data & semantic context

[Virtuoso](urn:osdi:sitevirtuoso/)

Relational tables, RDF Knowledge Graphs, and semantic data virtualization across databases, knowledge bases, filesystems, and APIs, without wholesale data duplication.

Intent-to-action orchestration

[OPAL](urn:osdi:siteopal/)

The harness that connects natural-language intent to executable queries and actions through loosely coupled Agents, Skills, and Tools, with cost-based LLM routing.

Persistent memory

[LOAM](urn:osdi:siteloam/)

Session histories, execution traces, entity relationships, decisions, and provenance, kept as reusable, machine-computable context.

The layers

## What each layer holds

Read top to bottom: people and agents ask, OPAL orchestrates, memory recalls, Virtuoso harmonizes, and the Data Spaces stay authoritative.

### AI Agents & Users

Who and what asks.

- **Human Users**Analysts, domain experts, knowledge workers
- **AI Agents**Claude, ChatGPT, Grok Bot, Muse, and others
- **Agent Frameworks**LangChain, AutoGen, CrewAI, custom harnesses
- **LLMs**GPT, Claude, Gemini, Grok, Muse, and others

### [OPAL](urn:osdi:siteopal/) — Orchestration Harness

Turns intent into governed action.

- **Natural Language Interface**Intent understanding and planning
- **Skills and Tool Routing**Loosely coupled Agents, Skills, Tools and Data Spaces
- **LLM Routing**Cost-based model selection
- **Policy and Governance**Identity, ABAC, guardrails and audit

### [LOAM](urn:osdi:siteloam/) — Entity Relationship Graph Collective

Linked Data memory that accumulates across sessions.

- **Session Graph**Conversations, plans, goals
- **Entity Graph**People, organizations, products, assets
- **Task Graph**Actions, tools, workflows
- **Trace Graph**Execution traces, provenance
- **Domain Graphs**Subject-specific knowledge (e.g., contracts, compliance)
- **Preference Graph**User and agent preferences
- **Additional graphs**Further domain-specific graphs

### [Virtuoso](urn:osdi:sitevirtuoso/) — Multi-Model DBMS

The semantic harmonization (virtualization) layer.

- **Data Federation**Connect and query across heterogeneous Data Spaces
- **Semantic Harmonization**Unified view using RDF, ontologies and Linked Data
- **Multi-Model Support**Relational, RDF, property graph and files
- **Standards-Based Access**SQL, SPARQL, GQL, GraphQL, HTTP, ODBC, JDBC, WebDAV, etc.

### Enterprise Data Spaces

Your data remains where it is.

- **Relational Databases**PostgreSQL, MySQL, SQL Server, Oracle, etc.
- **Knowledge Bases**Wikis, document repositories, content management systems, enterprise knowledge bases
- **Filesystems**Documents, PDFs, reports, logs, media
- **APIs**MCP tools, OpenAPI services, REST/SOAP, enterprise applications
- **Other Sources**IoT, data lakes, streaming data, etc.

Capabilities

## How the capabilities work together

The distinction isn’t any single feature. It’s how these capabilities work together without tight coupling to one database, one agent framework, one LLM, or one proprietary ecosystem.

| Capability | VOICE approach | Enterprise benefit |
| --- | --- | --- |
| Data access | Federated, standards-based access across existing Data Spaces | Less migration and duplication |
| Semantic context | RDF, Linked Data, ontologies, and semantic data virtualization | Consistent entity identification and relationship discovery |
| Agent orchestration | Loosely coupled Agents, Skills, Tools, and Data Spaces | Freedom to change agent frameworks and tools |
| LLM selection | Cost-based routing across supported models | Greater flexibility and cost control |
| Identity and governance | Verifiable identity ([YouID](urn:osdi:siteyouid/)) and fine-grained ABAC | Governed access for humans and agents |
| Persistent memory | RDF-based session history, relationships, and provenance | Reusable knowledge across interactions |
| Execution traces | Machine-computable records of relevant agent activities and outcomes | Auditability, debugging, and reproducibility |
| Progressive enrichment | New findings and relationships incorporated into persistent context | Knowledge that accumulates rather than disappearing between sessions |

A concrete example

## Enterprise contract intelligence

“Which customer contracts are affected by a supplier’s new compliance requirements, and what actions should we take?”

01 — Discover

Find what is relevant

Identify relevant contracts, suppliers, customers, and compliance requirements across databases, document repositories, and APIs.

02 — Connect

Correlate across sources

Use shared identifiers, RDF relationships, and ontologies to correlate information across those sources.

03 — Recall

Start from what is already known

Retrieve relevant findings, decisions, and execution traces from previous agent sessions.

04 — Reason

Pick the right model

Use an appropriate LLM alongside machine-computable context to evaluate potential impacts.

05 — Act

Governed actions only

Invoke authorized tools to prepare reports, create tasks, or initiate approval workflows, subject to access controls.

06 — Enrich

Keep what was learned

Preserve useful findings, new relationships, provenance, and execution outcomes for future interactions.

The next agent session doesn’t have to rediscover everything from scratch. Knowledge accumulates. Context improves. Actions remain governed.

How VOICE differs

## An architectural choice, not a replacement

VOICE complements the data, analytics, graph, and AI platforms an enterprise already runs, as well as offering an alternative to tightly integrated AI context stacks. Rather than making a particular platform the center of enterprise AI, it emphasizes six kinds of independence.

01

Data Space independence

Existing systems remain the authoritative sources.

02

Agent independence

Different agents and harnesses can use the same governed context.

03

LLM independence

Models can be selected or replaced according to task requirements and cost.

04

Standards-based interoperability

RDF, Linked Data, SQL, SPARQL, HTTP, ODBC, JDBC, MCP, and OpenAPI provide reusable integration interfaces.

05

Persistent, portable context

Machine-computable memory, provenance, and execution traces are not tied to one LLM’s conversation history.

06

Governance across boundaries

Identity, authentication, authorization, and storage remain distinct architectural concerns.

Business value

## What VOICE is designed to do for an enterprise

- **Reduce integration costs** by reusing existing systems and open standards.
- **Reduce AI operating costs** through selective LLM routing and reusable context.
- **Reduce repeated work** by preserving relevant session histories, execution traces, and accumulated knowledge.
- **Improve agent effectiveness** by combining current enterprise information with progressively enriched semantic context.
- **Strengthen governance** through fine-grained control over data access, memory retrieval, and action execution.
- **Protect existing investments** by avoiding unnecessary migrations and platform dependencies.

Most AI platforms help agents work with data. VOICE is designed to help agents work across Data Spaces, remember what they’ve learned, and execute governed actions using that knowledge.

## From intent to governed action

Context that grows more valuable over time.

[Talk to an Architect](urn:osdi:sitecontact/?dept=Solutions%20Architect) [Explore OPAL](urn:osdi:siteopal/)
