Question 1

How is Ontoxiom Harness different from an LLM Wiki?

Accepted Answer

An LLM Wiki has a model compile sources into markdown pages it maintains itself; agents read those pages to answer questions. It's great for capturing personal or team know-how.

But a wiki is a model-written summary. It can be stale, it can't count or enumerate precisely, it has no field-level permissions and it can't execute business actions. Agents in Harness work on typed objects continuously synced from your systems, so answers are exact down to the individual order. Every read and write passes the Governor, and actions go through checks, approvals and a full audit trail.

They can coexist: the wiki holds the “how”, the ontology holds the “what” and the “do”.

Question 2

We already have a RAG knowledge base. Do we still need an ontology?

Accepted Answer

RAG is good at policy and manual questions, but it retrieves similar text, not exact facts. Questions like “which orders are affected?” or “which customers fell more than 20% this month?” need queries, traversal and aggregation, and RAG can't do that.

An ontology doesn't replace your RAG: documents keep answering explanatory questions while the ontology handles facts and actions. In Ontoxiom both answer to the same permissions.

Question 3

How is an ontology different from GraphRAG or a knowledge graph?

Accepted Answer

GraphRAG uses a model to extract entities and relations from documents. The relations are inferred, and frozen at the last graph build. Traditional knowledge graphs are mostly used for search and visualisation.

Ontoxiom's ontology is an executable business model: types and relations are confirmed by the business and declared explicitly, data is synced continuously from your systems by Data Pipeline, and every write is validated. It also includes actions with preconditions and approvals, plus property-level permissions.

Question 4

Why not just let the LLM write SQL against our database?

Accepted Answer

It demos well but rarely survives production. The model has to guess cryptic table and column meanings, the generated SQL is hard to review, and above all a direct database connection means permissions depend on the prompt, with writes even less controllable.

In Ontoxiom, agents never see tables. They use eight semantic tools, and every call is authorised and audited by the Governor.

Question 5

How is this different from a data warehouse or data platform?

Accepted Answer

Warehouses and data platforms serve reporting and analytics; they organise tables and metrics. An ontology organises business objects, relations and actions, and serves the daily decisions and execution of people and agents.

They complement each other: your warehouse can be one of the ontology's sources, while the ontology gives AI a business layer it can understand and act on.

Question 6

Do we have to use Ontoxiom Harness?

Accepted Answer

No. Ontoxiom is open through MCP, REST and Python / Node SDKs. Popular agents such as WorkBuddy, Claude, ChatGPT and Cursor, or agents you build yourself, can connect to your enterprise ontology directly.

Harness is our ready-to-use workspace for companies that want frontline teams productive quickly. Whichever way you connect, permissions and audit are exactly the same.

Question 7

Which systems does the Harness desktop app run on? Can I use it on my phone?

Accepted Answer

The Ontoxiom Harness desktop app supports three platforms: Windows (64-bit, Windows 10 / 11), Mac with Apple chip, and Mac with Intel chip, on macOS 12 or later. The download page checks your device and recommends the right installer.

There is no iPhone, iPad, Android or Linux version today. If you open the download page on a phone, send the link to your computer and download there. If you'd rather work in another tool, connect Ontoxiom to the agent you already use over MCP or the SDK.

Question 8

Do we need to replace our ERP, MES or CRM?

Accepted Answer

No. Ontoxiom runs on top of your existing systems, connects to them through Data Pipeline and keeps them as systems of record. You can declare field by field which system is authoritative.

Question 9

Which LLMs are supported?

Accepted Answer

Use the model providers you choose, including open models self-hosted on your intranet. The ontology is decoupled from the model: when you upgrade or switch models, the business model and permissions don't need rebuilding.

Question 10

We have lots of PDFs, spreadsheets and scans. Can we use them?

Accepted Answer

Yes. Data Pipeline accepts PDFs (including scanned pages via OCR), Excel, CSV and pasted data. Models extract and normalise them against a declared schema, so new layouts need no bespoke training, and results can be previewed and reviewed before they're published to the ontology.

Question 11

Could an agent overstep its permissions or make a damaging change?

Accepted Answer

An agent never has more permission than the person it acts for. Every request passes the Governor first: deny wins, the default is deny, and permissions go down to individual properties.

Writes only happen through predefined actions: the effects are previewed and a check token is issued, anything needing approval goes to the Decisions inbox, and execution applies completely or not at all.

Question 12

Does our data leave our environment? Is it used for training?

Accepted Answer

It never has to. Ontoxiom supports private deployment and can use models you self-host. Qihuan never uses customer data to train any model.

Question 13

Can we deploy it privately?

Accepted Answer

Yes. The whole platform can run in your data centre, private cloud or dedicated cloud, including on an intranet.

Question 14

How do we know what the AI did?

Accepted Answer

Every request is recorded: who, through which agent, when, what was asked, which policy allowed or denied it, and the result. Every change to an object keeps its history, so you can see its state at any point in time.

Question 15

Is it safe to connect third-party agents like WorkBuddy or ChatGPT?

Accepted Answer

Third-party agents connect with personal API keys bound to a specific role and business scope. Their requests pass the same Governor and are written to the same audit log. They never get a database connection or system credentials, and can only do what's allowed.

If data must never leave your network, pair this with private deployment and self-hosted models.

Question 16

How soon will we see results?

Accepted Answer

The typical path is 1–2 weeks of discovery and a 4–8 week pilot. By the end of the pilot, the first high-value use case is running in production with measurable early results. Actual timing depends on systems, data quality and scope.

Question 17

Who maintains the ontology afterwards? What if the business changes?

Accepted Answer

The ontology is yours. Every change happens in a draft and is published as a reviewed, traceable version. During the Scale phase FDEs train your team to maintain and extend it independently, and we can keep supporting you.

Question 18

Is FDE just outsourced development?

Accepted Answer

No. FDEs are product engineers who work on site with your business team to model and launch. What they deliver is ontology, pipelines and agent capability on the platform, not one-off custom code. The goal is to transfer capability, not to create dependence.

Question 19

Which companies is this a good fit for?

Accepted Answer

Companies with several business systems (ERP, MES, WMS, CRM and so on), processes that cross departments, and a goal of having AI actually take part in execution, not just answer questions. Manufacturing, supply chain, cross-border trade and logistics are our most common industries.

Questions you might have.

Turn your business into a world AI can understand.