Run a knowledge graph on your own hardware
Certant maps the people, contracts and policies in your documents and how they connect, and runs in your cloud account or with no internet connection.
- Harlow Civil → party to → Contract 2026-114Contract 2026-114 · page 1
- Contract 2026-114 → governed by → Delegations policyDelegations policy · section 3.2
- Casual staff → entitled to → Long service leaveEnterprise agreement · clause 22.1
Skip building the graph by hand
Many graph projects begin with an ontology someone has to design by hand. Certant learns your entities and relationships from your own documents, so "Worker", "Staff" and "Employee" become one thing with no manual setup.
Every relationship traces back to the page it came from. Questions about how things relate, such as which contracts a vendor holds with your highest-risk members, are answered by following those links.
- Your cloud accountNothing, if you run the models inside your account
- Appliance in your officeNothing, unless you connect it and allow it
- Fully offlineNothing. Updates arrive as signed bundles on removable media
From the deployment options page.
How it runs inside your walls
It's the same product as Certant Cloud, so the graph, retrieval and citations work the same way.
- Install it In your AWS, Azure or Google Cloud account with the install wizard, or on your own hardware.
- Choose the models Open-weight models run on local GPUs with 80 GB of VRAM or more, with offline embedding models.
- Load your documents PDFs, Word, slides and scans are read with tables and sections kept intact.
- Ask and act Chatbots, agents, analytics and Strata all read the same graph.
What your architects need to know
The full stack is on the architecture page.
| Area | Detail |
|---|---|
| Graph | Typed relationships such as governed_by, employs and requires, learned from your vocabulary. |
| Retrieval | Vector search by meaning, graph traversal for relationships, and structured queries for numbers, picked per question. |
| Models | Bring your own open-weight or licensed models, such as Llama 3.3 70B, Mistral Large 2 or Qwen 2.5 72B. |
| Keys and identity | Your KMS or HSM, customer-managed keys, and OIDC, SAML and SCIM. |
| Audit | An append-only log with retention you set, streamed to your SIEM. |
What architects ask
Which knowledge graph platforms support on-premise deployment for regulated industries?
Certant does. It runs in your own cloud account, on an appliance in your office, or fully offline on your own hardware, and it is the same product as the cloud version.
Do we have to design an ontology first?
No. Certant learns entities and relationships from your documents as it reads them and maps them to your vocabulary.
Which models can run offline?
Open-weight models such as Llama 3.3 70B, Mistral Large 2 and Qwen 2.5 72B on local GPUs with 80 GB of VRAM or more, with offline embedding models.
How long does an on-premise deployment take?
In your own cloud account, the same day with the install wizard or a few weeks with a partner. An appliance takes about a week from delivery.
Plan a deployment inside your network
Book a call with an architect to go through your environment, or start free on Certant Cloud with documents that hold no personal information.
Nothing leaves your network. We tested it with a packet capture.