Rox Platform Architecture
Last updated
Last updated
Product Overview
PricingHow ROX Works
Platform ArchitectureCopyright Β© 2024 RoxAI. All rights reserved. 251 Rhode Island St, Suite 207,β¨San Francisco, CA 94103
The revenue stack has undergone a few key changes over the course of the last few years: (a) the shift from commit to consumption-driven revenue models, and (b) the proliferation of a wide array of point tools within the revenue ecosystem. These changes have resulted in revenue operations data moving from being structured single-tenant to multimodal, continuously evolving, and scattered across structured, unstructured, and semi-structured formats. This fragmentation necessitates a unified and flexible approach to unlock the full value of this data.
With the advent of Large Language Models (LLMs), there is a profound shift in the ability to extract structured information from unstructured and multi-modal data. They have fundamentally moved the abstractions higher to rethink every enterprise application. Compute abstraction has moved from containers to models while memory management has transitioned from traditional virtual memory to context window management.
In the context of the revenue stack, we believe that this paradigm shift redefines its core architecture. Storage moves from rigid relational databases to flexible data lakes capable of handling structured, unstructured, and semi-structured data. Computation moves from static, predefined business logic to reasoning engines that can dynamically interpret and act on multimodal data. Interfaces move from static dashboards to agentic, natural language-based systems accessible across desktop, mobile, and web. Together, these changes empower the revenue stack to meet the demands of modern, data-driven operations with unparalleled flexibility and intelligence. Rox is built with this shift as first class principles. In the next section, we will dive deep into Rox's Platform Architecture.
The architecture is split into 3 layers as explained above for the modern revenue stack -- the system of record, the compute and the interface layer.
The System of Record is an event-driven, unified knowledge graph built on multi-modal data that manages context for the AI Agents. All data from external systems or internal systems flow into the Rox Data Lakehouse. Once the data is in the lake house, Rox constructs a unified knowledge system across the multimodal data based on revenue domain entities. Access to this data is via event-driven APIs which the Agents use to read and write to the knowledge graph in near-real-time.
The Rox Data Lakehouse is multi-modal and processes data at TB+ scale. Unstructured data comes from in-house systems ingesting public sources like news, job postings, and blogs, while structured data comes from business systems such as CRM, Support, and product usage logs. For enterprises, Rox offers secure data sharing, ensuring no raw customer data is stored at rest. There is also an ETL option where Rox manages warehouse-native. For enterprise app data, Rox provides an API-based connector hub to integrate bi-directionally with apps like email, calendars, and Slack. More about the Data Lakehouse can be read HERE
After data enters the lakehouse, Rox builds a unified knowledge graph to map revenue entities like companies and people across distributed data sources. This in-house graph representation is continuously evolving to handle increasing enterprise data volumes. Streaming graph construction, which updates the graph in real-time as new data syncs, is a key focus area to manage rapid data ingestion efficiently. More about the Knowledge graph can be read HERE
Rox provides real-time, low-latency data access APIs that enable both read and write operations. These APIs power the reasoning engine and agentic interface while offering developer-friendly integration points for external systems. The data access layer also manages context for agents, ensuring the right data is retrieved for specific tasks. Enterprises can leverage these APIs to seamlessly interact with Rox's knowledge graph. More details are available HERE.
The Rox Agent Swarm operates as the central processing unit, bridging the System of Record (SOR) and Systems of Engagement. Its primary role is to execute both system- and user-driven actions. Each Agent within the Swarm is contextually scoped to a single sales representative and account pair, ensuring interactions are focused and highly personalized.
The Swarm architecture has three core components:
Natural Language Interface: Captures user intent and defines tasks.
Reasoning Pathways: Plans and executes tasks effectively.
Evaluation and Self-improvement: Continuously monitors and improves agent outputs.
The Natural Language Interface provides a seamless way for users to interact with the Agent Swarm through text or voice inputs. It translates ambiguous queries into actionable tasks using advanced natural language processing techniques like intent recognition, entity extraction, and context understanding. This ensures that user requirements are accurately interpreted and efficiently mapped to specific tasks.
Reasoning Pathways break down complex tasks into manageable actions. Using Rox Tools, agents identify and leverage the necessary public and private data sources to perform domain-specific actions. These tools are orchestrated in an optimized sequence during execution, ensuring tasks are completed effectively and efficiently.
The Evaluation System ensures continuous improvement of the Agentβs performance. It uses real-time monitoring, data labeling, and feedback to benchmark and optimize outputs. By aligning outputs with desired outcomes, the system ensures Rox Agents consistently deliver high-quality results tailored to the needs of each sales representative and account.
You can learn more about the agent swarm and the abilities it unlocks HERE
The Rox System agent has three key modes of engagement β mobile, web and native MacOS app. The system of engagement has three key objectives: meet sellers where they are, be one click away, and speak their language.
Meet sellers where they are
By offering multiple points of entry through web, mobile, and desktop applications, sellers can seamlessly engage with the system regardless of their preferred working environment. This flexibility ensures that the Rox Agent Swarm remains accessible and effective across different contexts and working styles.
Be one click away
This is made possible by the Rox agent swarm's omnipresent accessibility. During research phases, users can instantly summon the agent via a simple Command + K
shortcut, which triggers a chat window with the agent. In meetings, the native MacOS application enables quick agent deployment for real-time transcription and action item generation. For pre-meeting preparation, the Rox's audio readout capability allows for hands-free interaction, enabling users to multitask effectively while engaging with the swarm.
Speak their language
The system supports various natural, communication methods, seamlessly transitioning between text and voice interfaces. This flexibility in communication modes ensures that users can interact with the system in whatever manner best suits their current context or preference, whether they're typing queries, issuing voice commands, or receiving audio feedback. Rox obviates the need for complex UIs (i.e., workflow builders).