Frequently Asked Questions

What is a Knowledge Engine?

A software platform that codifies expert (often tacit) knowledge, semantically unifies data into a Knowledge Graph, and lets users (or agents) ask questions and trigger actions through natural-language prompts—democratising expertise without requiring users to be experts themselves.

How does Accrete define a Knowledge Function (KF)?

A KF is a specialised plug-in that runs inside the graph, applies domain algorithms (e.g., risk models, video vision), and creates new knowledge that downstream agents can chain together.

How is a Knowledge Engine different from an LLM or RAG system?

Unlike context-window prompt stuffing or RAG, Accrete stores entities + relationships explicitly, supports multi-hop reasoning, and persists new insights; LLM-only systems “reset each session” and miss hidden links.

What are the four layers of Accrete’s Knowledge Engine?

(1) Data Layer—ingests siloed sources; (2) Ingestion Layer—extracts & normalises subject-predicate-object triplets; (3) Knowledge Function Layer—runs codified expert models; (4) Agentic Layer—a suite of planners, memory, coding & other agents that invoke KFs and chat with users.

How does Accrete build an ontology automatically?

During a conversational onboarding, the agent asks clarifying questions, identifies entities/relationships in existing data, and links them into the graph—without teams of human ontology engineers.

Why are global ontologies hard, and how is this better?

Creating a single enterprise-wide schema is “expensive and time-consuming”; Accrete instead drives the ontology from the data itself, guided by the problem the user is trying to solve.

What makes Accrete’s graph unique vs. LLM context windows?

Accrete indexes relationships, enabling multi-step inference; LLM embeddings alone are “one-step or shallow” and lose structure.

How does persistent memory help?

Each new interaction is stored and compounds the graph for long-term learning, whereas vanilla LLMs forget once the prompt ends.

What is multi-hop reasoning in this context?

Agents can traverse several edges in the graph (e.g., company → supplier → foreign influence) to answer complex “why” or “impact” questions reliably, something embedding-only systems struggle with.

What roles do agentic Knowledge Functions play?

They expose APIs (e.g., narrative clustering, risk modelling) that planning agents orchestrate; coding agents even auto-generate execution plans.

How are Expert AI Agents related to the engine?

An Expert Agent is a composite that invokes one or more KFs plus the graph to deliver decision automation (e.g., ITSM Expert or Argus Supply-Chain).

Typical bootstrap steps for a new engine?

1️⃣ User describes a problem in chat → 2️⃣ Agent asks follow-ups → 3️⃣ Triplets extracted & normalised → 4️⃣ Graph built → 5️⃣ KFs added → 6️⃣ Agents deliver answers & dashboards.

What ROI did the ITSM Expert Agent achieve?

Ticket resolution dropped from 22 → 8 days for a Fortune-50 company, averting potential $1.2 M/hour downtime costs.

How does Nebula Social leverage the engine?

It encodes analyst know-how to predict viral narratives, cutting analyst “scroll time” 94 % and spotting influencers 6.7× faster.

Sales-pipeline automation example?

A Video Vision KF scores calls; agents blend those scores with CRM data to surface “top-3 deals at risk,” yielding 85 % time savings.

Can engines talk to each other?

Yes—Accrete exposes an interface so multiple engines can share selected knowledge, tools and agents.

What external sources can be integrated?

Foundation models, web-search agents, vector databases and any structured/un-structured corpora ingested through Layer 2.

How is security handled?

While granular security details are client-specific, data remain inside the customer’s graph; only authorised agents/KFs can access sensitive nodes—mirroring the platform’s focus on enterprise “ground truth.” (Implied throughout positioning docs.)

Does Accrete compete with model vendors like OpenAI?

No; model vendors supply raw language understanding, while Accrete delivers the reasoning & ground-truth layer those models lack.

How does Accrete capture tacit knowledge?

Expert interactions are logged in memory, then converted to graph facts and KF logic—scaling hard-won intuition across the org.

What’s next on the product roadmap?

Expansion of the multi-agent framework, more plug-and-play KFs, and deeper vertical solutions (e.g., finance, gov-intelligence) per internal roadmap slides.